Detection
of
epileptic
seizures
with
a
modified
heart
rate
variability
algorithm
based
on
Lorenz
plot
Jesper
Jeppesen
a,*
,
Sa´ndor
Beniczky
a,b,
Peter
Johansen
c,
Per
Sidenius
d,
Anders
Fuglsang-Frederiksen
aa
DepartmentofNeurophysiology,AarhusUniversityHospital,Noerrebrogade44,8000AarhusC,Denmark b
DepartmentofClinicalNeurophysiology,DanishEpilepsyCentre,VisbyA´lle5,4293Dianalund,Denmark c
DepartmentofEngineering,AarhusUniversity,Finlandsgade22,8200AarhusN,Denmark dDepartmentofNeurology,AarhusUniversityHospital,Noerrebrogade44,8000AarhusC,Denmark
1. Introduction
Theunexpectedandrandomoccurrenceofepilepticseizureis amongthemostdisablinganddangerousaspectsaffectingpatients withepilepsyandtheirfamilies.Notonlycanunattendedseizures leadtoserioustraumaticinjuriesandstatusepilepticus,butthey also increase the risk of sudden unexpected death in epilepsy (SUDEP).1
SUDEP isthe leadingcause of death in people withchronic uncontrolledepilepsy,2andrankssecondonlynexttostrokewhen
comparing years of potential life lost from SUDEP with other neurological diseases.3 Supervision hasprovedtobe associated
withasubstantialdecreaseinSUDEP.4InthiswaySUDEPtogether
with status epilepticus and traumatic accidents contributes epilepsy-related-deaths which couldbe prevented withseizure detectionalarmsystem.5Aseizurealarmsystemwouldgivethe
caregiverthepossibilityto:enableaproperposition(fromproneor supinetorecovery),stimulatethepatientbyrollingoverorsimilar means,cardio-pulmonaryresuscitationanddefibrillation, admin-ister medication or oxygen, clear the airway, protect against injuries,andseekhelpifneeded.5Theearlieraseizureisdetected thebetter.Secondarygeneralizedtonic–clonicseizuresdetected before the tonic–clonic phase would furthermore give the caregivers the possibility of securing the patient in a position where traumatic accidents could be avoided and prolong the windowofreactiontimeforthecaregivers.
ARTICLE INFO Articlehistory:
Received9October2014
Receivedinrevisedform6November2014 Accepted8November2014
Keywords:
Autonomicnervoussystem Epilepsy
Heartratevariability(HRV) Lorenzplot
Seizurealarm Seizuredetection
ABSTRACT
Purpose:Inordertoassesswhetherfocalepilepticseizurescanbedetectedanddistinguishedfrom exerciseweevaluated fourdifferentheartrate variability(HRV)methodswithshorttermmoving windowanalysisof30,50or100 R-Rintervalsorsecondsperanalyzedwindow.
Methods:Thefourmethodsconsistedof:(1)reciprocalhighfrequencypowerbasedonFastFourier Transformation,(2)CardiacSympatheticIndex(CSI),(3)ModifiedCSIbothbasedonLorenzplot,and(4) heartratedifferentialmethod.Seventeenpatients(12males,5females;age20–55)had47 seizures (includingthreesecondarygeneralizedtonic–clonic(sGTC)),whichwereanalyzedduringtheirlong termvideo-EEGmonitoringof1–5daysduration.Positiveseizuredetectionwasregarded,whenthe HRV-valueduringseizures(1minbeforeto3minafterseizure-onset)exceeded105%ofthehighest valueduringexerciseandnon-seizuresample-periodsofthesamepatient.
Results:ModifiedCSI100wasthemostaccuratemethod:itdetectedallseizuresfor13ofthe17patients within6sbeforetill50safterseizureonsettime,eventhoughexercisemaximumHRofeachpatient exceededthatoftheseizures.ThethreesGTCseizureswerealldetectedmorethanhalfaminutebefore thetonic–clonicphase.
Conclusion:Theresultsindicateadetectable,suddenandinordinateshifttowardsympatheticoverdrive inthesympathovagalbalanceoftheautonomicnervoussystemaroundseizure-onsettime,formost patients.TheModifiedCSIisapromisingparameterforaportableECG-basedepilepsyalarm,detecting bothfocalandsGTCseizures.
ß2014PublishedbyElsevierLtdonbehalfofBritishEpilepsyAssociation.
* Correspondingauthor.Tel.:+4578469909.
E-mailaddresses:[email protected](J.Jeppesen),[email protected]
(S.Beniczky),[email protected](P.Johansen),[email protected](P.Sidenius),
[email protected](A.Fuglsang-Frederiksen).
ContentslistsavailableatScienceDirect
Seizure
j o urn a lhom e pa g e :ww w . e l se v i e r. c om / l oca t e / y se i z
http://dx.doi.org/10.1016/j.seizure.2014.11.004
Portable epilepsy alarm systemsfor generalized tonic clonic (GTC)seizureshaveconsequentlybeeninvestigatedthoroughly.5 Differentmodalitiessuchaselectromyography(EMG),6
electroder-malresponse7andaccelerometry8havealreadybeenattemptedas
seizuredetectiondevicesforGTC-seizuresafterthe onsetofthe tonic–clonicphase;somehasalsobeencommercialized.5However,
aportableseizurealarmdetectingfocalseizuresandsecondary GTC-seizuresbeforethetonic–clonicphasehavenotyetbeenrealized.5
Heart rate variability (HRV) is a recognized method of estimating the parasympathetic and sympathetic tonus of the ANS.9Inter-ictalbaselinedifferencesofHRVbetweenpatientswith
epilepsyandhealthycontrolshavebeenstudiedextensivelyandin general suggest reduced HRV-parameters for patients with epilepsy.10–12 Also theHRand HRV changes frominter-ictal to
pre/post-ictal and ictal periods have, to some extent, been investigated.13,14In arecentlypublishedreview Egglestonetal.
reportedsignificantincreasesofheartrateduringseizuresfor82% ofpatientswithepilepsy.14Astudyexploringthepre-ictal HRV-changesin5minorlongerintervalsbeforedaytimeseizuresalso found an increasing HR, low-frequency/high-frequency power ratio(LF/HF),andCardiacSympatheticIndex(CSI),allindicating increasedsympathetic arousal, in thelast 5min preceding the seizurescomparedtoperiodsupto2hbeforetheseizure.15
Theheartratechangesoccurringduringseizuresarelikelyto happen due to the strong interconnecting between common epileptogenetic sites and the amygdalaand hippocampi which formspartofthecentral autonomicnervoussystem(ANS).16,17
Even if the site of epileptic origin is not part of the central autonomicnervoussystem,spreadofseizuresleadingto involve-mentofitbyictalactivityiscommon.17Alsoanteriorcingulate,
insular,posteriororbito-frontal,andthepre-frontalcorticesplay key roles in influencing the parasympathetic and sympathetic system14andspecificallyseizurepropagationtotheinsularcortex
havebeenshownbycorticalstimulationtoinfluenceheartrate.18
ThesechangesinthecentralANShaveadirectconnectiontothe ictalheartratechangesviathesympatheticandparasympathetic branchoftheANS.
HoweverthequestionremainsiftheANS(andHRV)changesin the peri-ictal period of seizures inducing tachycardia can be detected anddistinguishedfrom otherdaily changes known to influencetheANSandinducetachycardia,inparticularphysical exercise.AlthoughmostresearchregardingHRandHRVchangesin patientswithepilepsyhaveestimatedgroupdifferencesbetween epilepsypatientsandhealthycontrolsorbetweenictalphaseand non-ictalphase,afewstudieshavealso,withsomesuccess,triedto takeitonestepfurtherandproposedseizuredetectionalgorithms from ECG-based measurements.17,19–21 However, these studies havebeendonewhilepatientswereinrestingstatewithnocontrol periodsofexerciseorotherstrenuoustaskswhereheartratepulse is known to increase by which false positive alarmswould be suspected tooccur.17,19–22 Challenges of making an ECG-based
epilepsy alarm still include distinguishing arousal from sleep, standingfromreclining,anddailyexercisesfromseizures.5,17
The aimofthisstudy wastofindoutwhether acontinuous HRV-algorithm could reliably detect epileptic seizures and distinguishbetween thetachycardiarelatedtoseizuresand the oneduringexercisefromthesamepatient.Thehypothesis was thatselectedHRVparametersaremoresensitiveandspecificthan HRforthechangesrelatedtoseizureactivity.
2. Methods
2.1. PatientrecruitmentandECG-processing
Thirty-nineconsecutiveadultpatientswereincluded.Patients wereadmittedtothelongtermvideo-EEGmonitoringunit(EMU)
atAarhusUniversityHospital,DenmarkandattheDanishEpilepsy Center,Dianalund,Denmarkfordiagnosticreasonsorpre-surgical evaluation. Allpatients gave written,informedconsentand the studywasapprovedbytheEthicsCommittee(IDM-20110107).All patientswereaskedtoperformanexercisetest,whichconsistedof astepwisepulseincreaseoffirst110beats/minfor2min,second 140beats/minfor2min,andending withall-outmaximumfor 3min. The duration of the exercise test varied from 9 to 11min.TheECGwasrecordedusingleadII(rightclavicletoleft lowerribs(costae))withsamplingfrequencyof256or512Hzon thesamesystemasthevideo-EEG(NicoletOne)toensureoptimal time synchronization. The ECG-data of all the seizures were extractedfrom15to20minbeforeto10minafterseizure-onset,if noneoftheexclusioncriteria’sweremet.
ExclusioncriteriaforseizureHRV-analysis:
-Ifanytechnicalfeaturessuchasdisconnectionorexcessivenoise made theR-peak analysis impossible, thetime-length of the HRV-analysisfromtheseizure-sessionwasshortenedbutstill analyzed; however, if these technical features were present between3minbeforeto30safterseizure-onsettheseizurewas excluded.
-Seizureswereexcludedwhenanotherseizureoccurredlessthan 1hbeforetheseizureonset.
-Auras with no visible semiology and EEG-correlation were excluded.
-Ifpatientsdidnotcompleteatleaststep2(pulsestepof140bpm for2min)oftheexercisetestthepatientwasexcluded.
Outofthe39patients,sevendidnothaveanyseizures,nine werediagnosedwitheitherpsychogenicnon-epilepticseizuresor thediagnosiswasuncertain;onecouldnotperformtheexercise test because of knee problems; one had excessive ventricular prematurebeats;andfourcouldnotbeanalyzedduetoartifactsor othertechnicalproblems.Fortheremaining17patients(12males, 5females;age20–55,mean39years)allseizureonsetsandends wereestimatedwithprecisionof1sbytrainedexpertsblindedto the HRV analysis. If no EEG-correlation wasfound, but visible semiologywaspresent,thetimingofthefirstclinicalsemiological signofseizurewasusedasseizureonsetreference(usedforseven seizures,twopatients;nos.14and16).
For all17 patients withanalyzedseizuresthree non-seizure periodswere analyzed using same methods in orderto obtain baselinenon-seizurevaluestailoredtotheindividualpatients.The threenon-seizureperiodsconsistedof:
(1)Thefirst30mindaytimeawakenon-seizure periodwithout artifactsmorethan3hbeforethefirstseizure.
(2)A nighttime (sleeping) non-seizure period from 2.00am to 2.30am inthe nightbefore thefirst seizure(if patientwas awakethefirstfollowingsleepingperiodwasused).
(3)An 18–30min period containing the whole of the exercise session.1
Atotalofforty-sevenseizureswereincludedforHRV-analyses (seizureduration:range 20–1056s, median87s).TheECG-data fromall 47 extracted seizuresand all non-seizure periodswas further processed and analyzed in custom-made computer programs developed in LabVIEW 2011. Firstly ECG-data was prepared for R-peak detection using adjustable 5–15Hz finite impulse response high-passfilter to remove baseline drift and
1
Thenon-seizureselectedexercise-datawasifpossible30min,includingseveral minutesbeforeandafterexercise,butifsomedisconnectionorexcessivenoise artifactintheperiodoutsideexercisewaspresent,theperiodwasshortened.Stillall exercise-sessions,asaminimum,includedthewholeoftheexercise-test.
artifacts. Secondly automatic R-peak detection with manual editingensuredallR-peakwereselectedandfalsedetectedpeaks weredeletedbyanexperienceduser.
2.2. Movingwindowheartratevariabilityanalysismethods Four methods of HRV-analysis with three different time-lengthsofmovingwindowswereevaluatedforseizuredetection. 2.2.1. Reciprocalhighfrequency(HF)-power
The tachogram of all seizure and non-seizure periods were interpolated and resampled with 10Hz using cubic Hermite interpolationtosecureevensamplerateofthesignal.9Amoving
window of 30, 50 and 100swith maximum overlapping were applied and Fast Fourier Transformation (FFT) were used to computethePowerDensitySpectrumforeachwindow.Theband power of thehigh frequency power (HF)(0.15–0.4Hz) of each window wasthereafter extracted and thereciprocal HF-power computed. The values of each reciprocal HF-power from the movingwindowwereattachedtoa timeline,whereeachvalue depictedtheprior30,50and100sofreciprocalHF-powerofthe FFT-analysis.
2.2.2. Heartratedifferentialmethod
Theheartratedifferentialmethod(HR-diff)wasdonebysimple 2nd order central differential method, where the HR-diff was calculatedforallR-Rintervals(xi)andsummedtogetherinmoving windowswithmaximumoverlappingof30,50or100R-Rintervals forallperiodsanalyzed(1).
HR-diff¼X
k
i¼0
1
2ðxðiþ1Þxði1ÞÞ; k¼f30;50;100g (1) Thetimelinewasconstructedinsamewayasforthereciprocal HF-power except it was the prior 30, 50 or 100 R-R intervals insteadofsecondsthatrepresentedtheHF-diffvaluesatanygiven time.Thismethodsimplycomputesthepositiverateofheartrate changewithinthegivenwindow.
2.2.3. CardiacSympatheticIndex(CSI)andModifiedCSImethods Thesemethodshavebeendescribed morethoroughlyin our earlierpublication.23Briefly:theLorenzplot(orPoincareplot)is
usedtocalculatethe‘‘beat-to-beat’’R-Rintervalchangesgivenby fourtimesthestandarddeviationofthespreadinthetransverse axis(T-value)andthe‘‘overall’’R-Rintervalchangesgivenbyfour times the standard deviation of the spread deviation in the longitudinal axis (L-value)24 (Fig. 1), by which the CSI can be calculatedas(L/T)andModifiedCSIas(L2/T).Thesamemoving
windowsmethodof30,50and100R-RintervalsasfortheHF-diff methodwasusedtocalculatethecontinuousCSIandModifiedCSI values.Thetimelinewasalsoconstructedinthesamewaytoform acontinuousmeasurementofCSIandModifiedCSI(seeexamples ofModifiedCSIwith100R-RintervalsinFig.2).
2.3. Positiveseizuredetectionthreshold
Positiveseizuredetectionwasregardedwhenthevalueofthe specifiedHRV-methodwithina4-minwindowfrom1minbefore to3minafterseizure-onsettimesurpassed105%ofthemaximum of any non-seizure sample periods from the same patient (including exercise). The timing of the HRV-value passing the 105%thresholdlinewasregardedasdetectiontime.Thethreshold settingof105%wasappliedtocreateasafetymargintoenhance the reliability of distinguishing the non-seizure periods from seizuredetection.Themaximumofthevaluewithin1minbefore to 3min after seizure-onset time was also divided with the
maximum of thenon-seizure value during non-seizureperiods (includingexercise)forsamepatienttoestimatethereliabilityof thedetection.
3. Results
Of the four differenttypes of HRV-analyses,with the three different time-lengths of moving window, the Modified CSI100
detectedseizuresfrommostpatients(Figs.2and3).Thirteenoutof 17 patients had all seizures detected by the Modified CSI100
method,withmeandetectiontime16safterseizureonset(range: 6s before till 50s after seizure onset) (Table 1, Fig. 4). The 13patients (31seizures)withpositivedetectionusingModified CSI100 had seizure maximum values 1.93 times (mean; range:
1.31–3.19)greaterthannon-seizuremaximumvalue(Table2). Threeseizuresevolvedtobilateralconvulsiveseizures-sGTCS (patient 5, 4th seizure; patient 9, 4th seizure; patient 10, 1st seizure).AllthreeseizureshadModifiedCSI100detectionseveral
seconds beforethetonic–clonicphase (85s, 62sand35s).The ratios ofmaximumModifiedCSI100(seizurevs.non-seizure) for
theseseizureswereinrange1.59–2.00.
TheCSIandtheHR-diff methodsprovedusefulfordetecting seizures from some patients (CSI: 5–6 patients, HR-diff: 7–10 patients);howeverallofthesepatientsalsohadpositiveseizure detectionwiththeModifiedCSI100method(Table1).
ThereciprocalHF-powermethodwasnotabletodistinguish seizuresfrom exercise,as none of the47 seizureshad positive seizuredetectionforeither30,50or100sofmovingwindowing. Theexercisemaximumpulseexceededtheseizuremaximum pulseforallpatients.
4. Discussion
In search for an optimal seizure detection algorithm, we analyzed four differentHRV-methods each of which we tested with three differenttime-resolutions of moving windows. Our mainaimwastodistinguishtheheartratechangesseenduring exercisefromthoseduringseizuresusinganECG-basedseizure detectionalgorithm.
Thegeometricandnon-stationaryLorenzplotmethodproved tobethebestmethodofexpressingthefastinvolvingchangesin theautonomicnervoussystem(ANS)duringseizures.Theresults demonstrating13outof17patientshavingpositivedetectionsfor all seizures with Modified CSI100 likewise indicate that this
measureableANSchangeexceedsandcanbedistinguishedfrom
Fig.1.Lorenzplot:examplefromnon-seizurerestpatient7,with50R-Rintervals. The transverse axis (T) reflects beat-to-beat variation (T=4SD1), while the longitudinalaxis(L)reflectstheoverallfluctuations(L=4SD2).
exerciseformostepilepticpatientswithfocalseizures.Ageneral group-differencebetweenCSIininter-ictalandictalperiodshave also been shown in another study,13 however a continues
algorithmbasedonLorenzplotwithindividualthresholdsetting forseizuredetectionhavetoourknowledgenotbeeninvestigated before.
We hypothesize that the pathophysiological mechanisms behindtheModified CSIalgorithm arerelated totheabnormal increase in sympathetic tonus during seizures. A standardized pacedbreathing study of healthy subjectsshowed theSD2 (L) componentoftheLorenz/Poincareplottobestronglycorrelated with the recognized sympathetic expresser of the frequency domain,LF/HF.25Thissupportsthemathematicalperceptionofthe
proposedModifiedCSIalgorithm(L2/T)beingstronglyrelatedwith
thesympathetictonusoftheANS,asitemphasizestheLvaluetoa
greaterextent thantheCSI algorithm (L/T).TheCSI-value (and mostlikelyalsotheModifiedCSI-value)mirrorsthesympathetic tonusoveraperiodoftimeandthusourresultsindicateasudden andexcessivesympatheticshiftinthesympathovagalbalanceof theautonomicnervoussystem(ANS)justaroundseizure-onsetfor mostpatientswithepilepsy.
AsreportedinanotherstudyusingECG-detectionalgorithms17
wealsodidnotfindanyapparentcorrelationbetweenthelocation oftheepilepticfocusandpositivedetection.Thepatient-specificity ofseizuredetection,whichourresultshaveindicated,havealso beendescribedinotherECG-basedseizuredetectionstudies17,20,26
and patient specific HR patterns acrossdifferent seizureshave likewise been reported.16 Also, Epstein et al. have found an
additionalincreaseinmeanheartratewhenseizuresspreadfrom regionallimbictobilateralcorticalareas.27Therefore,onecould Fig.2.ModifiedCSIcontinuouscalculatedusingmovingwindowsof100R-Rintervalsexample.(A)Patient7,2ndseizure,graycolorrepresentstimeofseizure.(B)Lorenzplot sampleduringtimeofseizureonset.(C)Patient7,exercisetest.(D)Lorenzplotsampleduringexercisetest.
speculate that theictal HR patternsare likely tobe related to specificpatternsofseizurespread.Thus,weassumethatseizure detections werehighly patient-specific due tothesimilarity of specificfocalseizurepropagationandspreadinthebrainwithin, but not between, patients. Thereby we also appreciate that a genericECG-basedseizuredetectionalarm(thatwouldworkforall patientswithepilepsy)doesnotseemplausiblewiththedetection
methodsandalgorithmspresentedhere.However,distinguishing thepatients,forwhomourproposedseizurealarmsystemwould bepossible,couldsimplybedonewithsemi-automaticanalysesof ECG-datafromlongtermvideo-EEGmonitoringwithseizuredata orseizuresandnon-seizureperiodsrecordedwithECGbyother means. We thereby suggest a patient-tailored threshold and personalized detection setting giving a high specificity and
Table 1
SeizuredetectionwithCSI,ModifiedCSIandHR-differentialmethods.
sensitivityoftheseizuredetectingsystem,whichalsohavebeen proposedbyotherstudies.14,20,26
Three of the analyzed seizures evolved to sGTCS. All were detectedusingModifiedCSI100,morethanhalfaminutebeforethe
tonic–clonicphase.AlthoughbasedonasmallsamplesizeofsGTCS inthisstudy,analgorithmdetectingthesGTCSbymorethanhalfa minutebeforethetonic–clonicphasewouldhavegreatimportance forcaretakerswhensupportingthepatient.Furthermore,itraises thepossibilityofterminatingthemoredangerousanddamaging convulsivephaseoftheseizuree.g.withfastactingmedicineor on-demandvagus nervous stimulation(VNS).14A portable seizure
detectionsystemalarmingrelativesorotherstotakeactionbefore orduringemergenciessuchasprolonged-,repetitive-, secondary-generalized tonic clonic seizures or status epilepticus could in somecasesalsohelppreventsuddenunexpecteddeathinepilepsy (SUDEP).5ECG-baseddetectiondevicescouldincluderesponsive
stimulationofheart,respirationormuscles(including diaphrag-maticpacing),administrationofmedicationoroxygen.5Areliable
seizuredetectionsystemwithobjectiveseizurelogswouldalso enhancethepatient-diarytechnique,whichissubjectiveandcan beinfluencedbyconsciousnessormemory.5,22
Expressingheartratevariability parametersbymeansofFFT powerbandshavebeenpopularandareoftenusedasstandard measurementswhendoingHRV-analysisof5minorlonger.9Ina
pilot-studywe likewise sawpromisingresultsof usingthe FFT analysismethodasseizuredetectorbydoingveryshortmoving windowFFTanalysisof64R-Rintervalswhere6of6seizuresfrom 3patientsshowedhigherreciprocalHF-powerjustaroundseizure onsetcomparedtonon-seizuresampleperiods.21However,inthe
pilotstudywedidnotincludeacontrolperiodofexerciseforthe patients,whichprovedtohaveagreatinfluenceontheresults,as seeninthepresentstudy,wherenoneofthe47seizuresanalyzed had higher reciprocal HF-power than the patient’s own non-seizure periods (exercise). This strongly indicates that the reciprocalHF-power toa large extentfollow the degreeof the maximumpulserate, which wasboth higherand had a longer durationfortheexerciseteststhananyseizure,forallanalyzed patients.
To the best of our knowledge,this is the first study which comparesHRV-changesduringseizuresandexercise.Wechoose thestepwise pulse bike-exercise test in orderto standardize a physicalexercisewhichwe weresurethat mostpatients could complete.Itcould,however,infuturestudiesalsobeinterestingto testifthepositiveresultswouldendureiftestedagainstotherdaily exerciserelatedactivitiessuchasclimbingstairs,liftingbagsand otherpromptphysical activation,inwhich a suddenheart rate raisemaybeexpected.5
We compared the seizure period maximum values of the differentHRVdetectionmethodswiththreenon-seizureperiodsof ½-hHRV-data.Thelimitationofthenon-seizureperiodswasset
because of the time-consuming task of data processing and manuallyensuring correctR-peakdetection.However, thehigh individualdetectionsensitivityof100%for13of17patientsusing theModifiedCSI100algorithm,andthefactthatthehighest
non-seizurevaluesoftenwereseenduringexercise,webelieveproves the realistic perspectives in using Modified CSI algorithm as a significantparameterinpersonalizedECG-basedseizuredetection alarm for a majority of patients with epilepsy. This is to be challengedin ablinded prospectivestudywhere thesensitivity andspecificityofthismethodwillbeevaluatedandlonger non-seizureperiodswillbeusedascontrol.Furthermore,suchastudy mightalsotrytoidentifythedifferencebetweenthepatientswith andwithoutpositiveseizuredetectioninregardstotheseizures focalonsetandpossiblespreadtotheANSpathways(forinstance byusingMagnetoencephalographyandEEGsourceimaging).
5. Conclusion
TheresultsindicatethataportableECGseizuredetectionalarm system distinguishing the heart rate tachycardia seen during seizures withthat during exerciseseems highly possible for a majorityofpatientswithepilepsybasedonthemethodspresented here.TheModifiedCSIdistinguishedaccuratelybetween physio-logic,exercise-inducedECG changesand theictalECG changes, whileasimpleanalysisoftheHRfailedtodoso.Althoughsome patients do not develop specific detectable heart ratechanges duringseizures,itcaneasilybedecidedforwhichpatientsan ECG-seizure-alarmsystemwouldworkbyanalyzingpatientECG-data fromvideo-EEGmonitoringunitsorbyothermeans.
Conflictofinterest
We have no conflicts of interest or financial support with regardstotheauthorshiporpublicationofthisarticle.
Acknowledgements
We would liketoacknowledge theEMU-staffof theDanish Epilepsy Center and the Department of Neurology, Aarhus University Hospital for helping with the special settings and recordingofthepatients,andanadditionallyappreciationtonurse BirgitteHannerforalsohelpinginorganizingtherecruitmentof thepatients.
TheresearchwasfinancialsupportedbyTheDanishCouncilfor IndependentResearch(11-113323),Lundbeck Foundation (R83-A7684), Aarhus University (2010-218/2-82), Aase og Ejner DanielsensFoundation(10-000387),andBrd.Hartmann Founda-tion(A14291).Thefinancialsupportershadnoinvolvementinany partoftheresearch.
Table 2
Seizuredetectionmeantimeandseizuremax.vs.non-seizuremax. HRV-method Numberofpatientsallseizures
havepositivedetection
Meantimeto105%thresholdlinefromseizureonset(s)a
Ratioofmeanseizuremax.vs. non-seizuremaxa CSI30 6 14 1.61 CSI50 7 12 1.90 CSI100 5 18 1.66 Modif.CSI30 3 25 1.62 Modif.CSI50 7 12 2.01 Modif.CSI100 13 16 1.93 HR-diff30 7 7 1.62 HR-diff50 10 6 1.58 HR-diff100 9 10 1.68 a
Calculatedonlyforpatientswithpositivedetectionforallseizuresforthegivenmethod.
References
1.ShorvonS,TomsonT.Suddenunexpecteddeathinepilepsy.Lancet2011;378: 2028–38.
2.TomsonT,NashefL,RyvlinP.Suddenunexpecteddeathinepilepsy:current knowledgeandfuturedirections.LancetNeurol2008;7:1021–31.
3.ThurmanDJ,HesdorfferDC,FrenchJA.Suddenunexpecteddeathinepilepsy: assessingthepublichealthburden.Epilepsia2014;55:1–7.
4.Langan Y, Nashef L, Sander JW. Case–control study of SUDEP. Neurology
2005;64:1131–3.
5.VandeVelA,CuppensK,BonroyB,MilosevicM,JansenK,VanHuffelS,etal.
Non-EEGseizure-detectionsystemsandpotentialSUDEPprevention:stateof theart.Seizure2013;22:345–55.
6.ConradsenI,BeniczkyS,WolfP,JennumP,SorensenHBD.Evaluationofnovel algorithmembeddedinawearablesEMGdeviceforseizuredetection.ConfProc IEEEEngMedBiolSoc2012;2012:2048–51.
7.PohM-Z,LoddenkemperT,ReinsbergerC,SwensonNC,GoyalS,SabtalaMC,
etal.Convulsiveseizuredetectionusingawrist-wornelectrodermalactivity andaccelerometrybiosensor.Epilepsia2012;53:e93–7.
8.BeniczkyS,PolsterT,KjaerTW,HjalgrimH.Detectionofgeneralizedtonic– clonicseizuresbyawirelesswristaccelerometer:aprospective,multicenter study.Epilepsia2013;54:e58–61.
9.MalikM,BiggerJ,CammA,KleigerRE,MallianiA,MossAJ,etal.Heartrate variabilitystandardsofmeasurement,physiologicalinterpretation,andclinical use.EurHeartJ1996;17:354–81.
10.TomsonT,EricsonM,IhrmanC,LindbladLE.Heartratevariabilityinpatients withepilepsy.EpilepsyRes1998;30:77–83.
11.AnsakorpiH,KorpelainenJT,HuikuriHV,TolonenU,Myllyla¨ VV,Isoja¨rviJIT. Heartratedynamicsinrefractoryandwellcontrolledtemporallobeepilepsy.J NeurolNeurosurgPsychiatry2002;72:26–30.
12.LotufoPA,ValiengoL,Bensen˜ orIM,BrunoniAR.Asystematicreviewand meta-analysisofheartratevariabilityinepilepsyandantiepilepticdrugs.Epilepsia
2012;53:272–82.
13.PonnusamyA,MarquesJLB,ReuberM.Comparisonofheartratevariability parametersduring complex partialseizures andpsychogenicnonepileptic seizures.Epilepsia2012;53:1314–21.
14.EgglestonKS,OlinBD,FisherRS.Ictaltachycardia:thehead–heartconnection.
Seizure2014;23:496–505.
15.BehbahaniS,DabanlooNJ,NasrabadiAM,TeixeiraCA,DouradoA.Pre-ictal heartratevariabilityassessmentofepilepticseizuresbymeansoflinearand non-linearanalyses.AnadoluKardiyolDerg2013;13:797–803.
16.Leutmezer F, Schernthaner C, Lurger S, Po¨tzelberger K, Baumgartner C. Electrocardiographic changes at the onset of epileptic seizures. Epilepsia
2003;44:348–54.
17.OsorioI.AutomatedseizuredetectionusingEKG.IntJNeuralSyst2014;24: 1450001.
18.OppenheimerSM,GelbA,GirvinJP,HachinskiVC.Cardiovasculareffectsof humaninsularcortexstimulation.Neurology1992;42:1727.
19.JansenK,VaronC,VanHuffelS,LagaeL.Peri-ictalECGchangesinchildhood epilepsy:implicationsfordetectionsystems.EpilepsyBehav2013;29:72–6.
20.KeremDH,GevaAB.Forecastingepilepsyfromtheheartratesignal.MedBiol EngComput2005;43:230–9.
21.JeppesenJ,BeniczkyS,Fuglsang-FrederiksenA,SideniusP,JasemianY. Detec-tionofepileptic-seizuresbymeansofpowerspectrumanalysisofheartrate variability:apilotstudy.TechnolHealCare2010;18:417–26.
22.OsorioI,SchachterS.Extracerebraldetectionofseizures:anewerain epilep-tology?EpilepsyBehav2011;22(1):S82–7.
23.Jeppesen J, Beniczky S, Johansen P, Sidenius P, Fuglsang-frederiksen A. UsingLorenzplotandCardiacSympatheticIndexofheartratevariabilityfor detectingseizuresforpatientswithepilepsy.ConfProcIEEEEngMedBiolSoc
2014:4563–6.
24.ToichiM,SugiuraT,MuraiT,SengokuA.Anewmethodofassessingcardiac autonomicfunctionanditscomparisonwithspectralanalysisandcoefficientof variationofR-Rinterval.JAutonNervSyst1997;62:79–84.
25.GuzikP,PiskorskiJ,KrauzeT,SchneiderR,WesselingKH,WykretowiczA,
etal.CorrelationsbetweenthePoincare´ plotandconventionalheartrate variability parameters assessed during paced breathing. J Physiol Sci
2007;57:63–71.
26.VanElmptWJC,NijsenTME,GriepPAM,ArendsJBAM.Amodelofheartrate changestodetectseizuresinsevereepilepsy.Seizure2006;15:366–75.
27.EpsteinMA,SperlingMR,O’ConnorMJ.Cardiacrhythmduringtemporallobe seizures.Neurology1992;42:50.