Reliable Clinical Monitoring using Wireless Sensor ...lu/talks/sensys10-hospital.pdf · Reliable Clinical Monitoring using Wireless Sensor Networks: Experiences from a Step ... pulse

ReliableClinicalMonitoringusingWirelessSensorNetworks:ExperiencesfromaStep‐downHospitalUnit

OctavChipara,ChenyangLuThomasBailey*,Gruia‐CatalinRomanComputerScience&Engineering,SchoolofEngineering

*DepartmentofMedicine,SchoolofMedicine

MoAvaAon

  ClinicaldeterioraDoninhospitalizedpaDents  4‐17%sufferfromadverseeventssuchascardiacorrespiratoryarrest.

  Upto70%ofsucheventscouldhavebeenprevented.

  EarlydetecDonofclinicaldeterioraDonbasedonvitalsigns  clinicaldeterioraDonisoOenprecededbychangesinvitals

  Real‐DmepaDentmonitoringisrequired  wiredpaDentmonitoringequipmentinIntensiveCareUnits  mostgeneralhospitalunitscollectvitalsmanually  wirelesstelemetrysystemstooexpensiveforwideadopDon

2

Goal:reliablewirelessclinicalmonitoringforgeneralhospitalunits

OurApproach

1.  Buildaclinicalmonitoringsystemusingsensornetworks.2.  Deployitinageneralhospitalunitover7months.

3.  Clinicaltrialwith46hospitalizedpaDents.

4.  HolisDcsystemreliabilitystudy:networkandsensor.

5.  DemonstratepotenDalforclinicaleventdetecDon.

3

SystemArchitecture

  BasestaDon•  LaptopconnectedtoWi‐Fi

  Relays•  Pluggedintowalloutlets•  Redundantdeployment

  coverage  faulttolerance

  Portablepulseoximeter•  pulseoximeter+microcontroller+

radio

•  ba`eryoperated

4

ReliableNetworkArchitecture

Problem:PaDentsingeneralhospitalunitsareambulatory.SoluDon:Two‐Derarchitectureforend‐to‐enddatadelivery.

1  DynamicRelayAssociaDonProtocol(DRAP):PaDent‐>1strelay  DynamicallyassociatethepaDentnodewitharelay

  Single‐hopprotocolhandlespaDentmobility  SimplifypowermanagementinpaDentnodes(sendonly)

2  StaDonaryrelaynetwork:1strelay‐>…‐>basestaDon  ReusewelltestedmeshrouDngprotocol:CTP

  IsolatedfrompaDentmobility

  Wall‐plugged=>noneedtoworryaboutenergy

5

ClinicalDeployment

  Step‐downcardiaccareunit•  16paDentrooms,1200m2

  Network•  18relays:redundantnetwork•  Longestpath:3‐4hops•  Channel26ofIEEE802.15.4

  PulseandoxygenaDoncollectedevery30or60seconds

  46paDentsenrolled•  >41daysofmonitoring

•  2‐68hoursperpaDent•  5paDentsexcludedfromanalysis•  Upto3paDentsataDme

6

PotenAalforDetecAngClinicalDeterioraAon

!"

#$%&

7

Bradycardia

!"

#$%&

Pulmonaryedema

!"

#$%&

Sleepapnea

SystemReliability

  Networkreliability>95%forallpaDents.  EffecDvenessofDRAP+CTP

  Mediansensingreliability>80%.  But29%ofpaDentswithsensingreliability<50%

  Systemreliabilitydominatedbysensingreliability.

8

NetworkReliability  Time‐to‐failure

  TimeintervalduringwhichasystemconDnuouslyoperateDllafailureoccurs.

  Time‐to‐recovery  TimeintervalfromtheoccurrenceofafailureDllwhenthesystemrecovers.

9

Recoverfrom90%offailureswithin2min=>Quickrecoveryfromfailure

CDFofAme‐to‐failure CDFofAme‐to‐recovery

MedianCmetofailure=19min

SensingReliability  Failuresarecommon:medianDme‐to‐failure<2min

  Recoverfrom90%ofsensingfailureswithin4min  Transientfailurescausedbyhumanmovement

  Long‐taileddistribuDonforDme‐to‐recovery  SensordisconnecDon

10

90%ofoutages<4min

Longtail

CDFofAme‐to‐failure CDFofAme‐to‐recovery

RelaxSamplingRates

  Increaserequiredsamplingperiodto5,10,15min  Oversampleat1‐2reading/min  Consideredasuccessifonevalidmeasurementperrequiredsamplingperiod

  SDllordersofmagnitudehigherratethanmanualmeasurement

  Higherreliabilityatasamplingperiodof5min  Diminishingreturnatlongersamplingperiods

11

Sensingreliabilityatdifferentsamplingrates

SensorDisconnecAonAlarms

  AutomaDcallynoDfynurseaOerreceivingnodataforaDmeoutthreshold

  15minDmeoutbalances#alarmsvs.reliabilitygain  Infrequentalarms:1.55intervenDonsperpaDent,perday  Similarreliabilityto5and10minDmeouts

12

SensingreliabilitywithdifferentAmeouts #alarmsperday

PuLngthemtogether

  DisconnecDonalarmsandoversamplingarecomplementary.•  DisconnecDonalarms=>handlessensordisconnecDon•  Oversampling=>handlesintermi`entfailurescausedbymovement

  88%ofpaDentswith>70%sensingreliabilitywith5minsamplingperiodand15minDmeout

13

WirelessSensorNetworksvs.Wi‐Fi

  MoreenergyefficientthanWi‐Fiatlowdatarate  Commonvitalsignshavelowdatarate.  Nursesaretoobusytochangeba`eries!

  Lowdeploymentcost  Meshnetworkswithoutwiredinfrastructure.

  EaseadopDon(e.g.,fieldhospitals,ruralareas).  EvenmajorhospitalsmaynotguaranteefullWi‐Ficoverage.

  Sufficientreliability  Mediannetworkreliability>99%inourclinicaltrial.  Evenawirednetworkwouldimprovereliabilityonlymarginally.

14

Relatedprojects

  Assistedliving:ALARM‐NET

  Disasterrecovery:AID‐N,CodeBlue,WIISARD

  Emergencyroom:MEDISN,SMART

  MoDonanalysis:Mercury

  Commercialwirelesstelemetry(Phillips,Cisco,GE)•  WiFi=>single‐hopwireless,wiredbackbone•  adopDonlimitedtospecializedhospitalunits

15

Conclusion

  Wirelessclinicalmonitoringforgeneralhospitalunits.  Clinicaltrialinastep‐downhospitalunit

•  Highlyreliablenetwork

•  Systemreliabilitydominatedbypulseoximetersensors  Oversampling

  DisconnecDonalarms

•  PotenDalfordetecDngclinicaldeterioraDon

  On‐going:real‐DmeclinicaleventdetecDon  IntegraDonwithelectronicmedicalrecords  EventdetecDonbasedonmachinelearning

  AutomaDcalarmsforearlyintervenDon  LargerclinicaltrialofeventdetecDonsystem

16

Acknowledgement

  NursesatBarnesandJewishHospital

  NIHNaDonalCenterForResearchResources:CTSAgrant

  Barnes‐JewishHospitalFoundaDon

  NSFNeTS‐NOSS,CRIgrants

17