neurocritical care program uc sf critical care bioinformatics at ucsf j. claude hemphill iii, md,...
Post on 20-Dec-2015
221 views
TRANSCRIPT
NEUROCRITICAL CARE PROGRAM
UCSF
Critical Care Bioinformatics Critical Care Bioinformatics at UCSFat UCSF
J. Claude Hemphill III, MD, MASJ. Claude Hemphill III, MD, MAS
Kenneth Rainin Chair in Neurocritical CareKenneth Rainin Chair in Neurocritical CareAssociate Professor of Clinical Neurology Associate Professor of Clinical Neurology
and Neurological Surgeryand Neurological SurgeryUniversity of California, San FranciscoUniversity of California, San Francisco
Director, Neurocritical CareDirector, Neurocritical CareSan Francisco General HospitalSan Francisco General Hospital
Disclosures Research Support: NIH/NINDSConsulting: UCB Pharma Stock (options): Cardium Therapeutics (Innercool Therapies), Ornim
So What’s the Problem?So What’s the Problem?
• Some of what we don’t knowSome of what we don’t know
1)1) Do secondary brain insults have a dose-response Do secondary brain insults have a dose-response relationship with outcome?relationship with outcome?
2)2) We treat univariate in a multivariate worldWe treat univariate in a multivariate world Interaction and relationship between various Interaction and relationship between various
physiologic parameters?physiologic parameters? Event signatures?Event signatures?
3)3) How do we integrate new measures (e.g. PHow do we integrate new measures (e.g. PbtbtOO22)?)?
4)4) How often do we need to collect physiologic data How often do we need to collect physiologic data to optimize patient care?to optimize patient care?
This is complicated
Looking at ICU Data BedsideLooking at ICU Data Bedside
Paper charts in most ICUs, electronic charts in some
ICUInformatics
2009
Neurocritical Care Database/Informatics
GOALSGOALS
1)1) Identify physiological signatures to diagnose patients and predict Identify physiological signatures to diagnose patients and predict outcomesoutcomes
2)2) Use real-time data to rationally drive clinical decisions and treatment based Use real-time data to rationally drive clinical decisions and treatment based on the specific physiologic abnormalityon the specific physiologic abnormality
3)3) Determine dosage and delivery for commonly used NICU medicationsDetermine dosage and delivery for commonly used NICU medications
4)4) Suggest new clinically-relevant experimental research modelsSuggest new clinically-relevant experimental research models
5)5) Develop user-friendly “behind the scenes” data analysis that aids Develop user-friendly “behind the scenes” data analysis that aids interpretability and clinical applicability interpretability and clinical applicability
UCSF Approach to Critical Care InformaticsUCSF Approach to Critical Care Informatics
• Centered at SFGHCentered at SFGH– Trauma CenterTrauma Center– Stroke CenterStroke Center
• Driven by interest of specific cliniciansDriven by interest of specific clinicians– Claude Hemphill, MD,MAS - neurointensivistClaude Hemphill, MD,MAS - neurointensivist– Geoff Manley, MD,PhD - neurosurgeonGeoff Manley, MD,PhD - neurosurgeon– Mitch Cohen, MD – trauma surgeonMitch Cohen, MD – trauma surgeon
• Focus on neurotraumaFocus on neurotrauma• ““Ground up” approachGround up” approach
– Develop infrastructureDevelop infrastructure– Knowledge discovery (research driven)Knowledge discovery (research driven)– Not trying to feed back immediately into Not trying to feed back immediately into
clinical care – too earlyclinical care – too early
UCSF Initial EffortsUCSF Initial Efforts
• Gather some dataGather some data– Kiosk methodKiosk method– ““Home grown” softwareHome grown” software
• Analyze in novel, but simple waysAnalyze in novel, but simple ways– Detection of secondary brain insultsDetection of secondary brain insults– Improved univariate measures – Improved univariate measures –
AUC (area under the curve)AUC (area under the curve)
• Identify and engage collaborators with Identify and engage collaborators with expertise (generally not clinicians)expertise (generally not clinicians)
• PublishPublish
NICU Data Acquisition 2003NICU Data Acquisition 2003
• Independent CPUIndependent CPU
• Multiple serial portsMultiple serial ports– Overhead monitor Overhead monitor
(Philips)(Philips)
– Ventilator (Draeger)Ventilator (Draeger)
– Brain OBrain O22 (Licox) (Licox)
– CBF (Hemedex)CBF (Hemedex)
• Data time-synchedData time-synched
• Operator must initiate Operator must initiate data acquisitiondata acquisition
How Often Do We Need to Collect this Data?How Often Do We Need to Collect this Data?
• Current standardCurrent standard– Paper chart - Q 1 hour Paper chart - Q 1 hour
and as neededand as needed
– CareVue (electronic CareVue (electronic medical record) – medical record) – up to Q 15 minup to Q 15 min
• Study comparing Q 1 min Study comparing Q 1 min v. medical record (MR) for v. medical record (MR) for SBI identification and SBI identification and dose (n=16; 72 hours dose (n=16; 72 hours each)each)
ICP > 20
Subject# of Events AUC in mmHg.min
Q 1 min MR Q 1 min MR
1 1 1 0.5 0.1
2 10 6 13.8 9.8
3 1 0 6.1 0
6 2 10 3.0 3.0
7 9 5 76.6 73.7
8 0 1 0 0.1
9 0 1 0 1.5
10 21 12 22.1 25.6
11 0 0 0 0
12 0 14 0 25.9
13 0 0 0 0
14 7 76 4.3 33.6
15 1 11 0.4 8.5
16 4 4 7.6 13.5
17 40 23 59.9 73.7
Hemphill, Physiological Measurement, 2005
Borrowing from PharmacokineticsBorrowing from Pharmacokinetics
• ““Dose” is area under the curve (AUC)Dose” is area under the curve (AUC)
35
35.5
36
36.5
37
37.5
38
38.5
39
39.5
0 20 40 60 80 100 120 140 160
Hours from Hospital Admission
Bo
dy
Tem
per
atu
re
+
Does It Matter How we Define Dose?Does It Matter How we Define Dose?
SBI Odds Ratio 95% CI P
Any hypotension (n=26) 3.39 1.34-8.56 .009
1 episode of hypotension 2.05 0.67-6.23 0.21
≥ 2 episodes of hypotension 8.07 1.63-39.9 0.01
Minimal dose hypotension(< 1 mmHg*minute)
1.35 0.28-6.4 0.71
Moderate dose hypotension (1-100 mmHg*minutes)
3.14 0.85-11.6 0.087
High dose hypotension(> 100 mmHg*minutes)
12.55 1.5-107 0.021
Impact of ED episodes and dose of hypotension on risk of in-hospital death after severe TBI (n=107)
*
* Manley, Arch Surg, 2001 +Barton, Acad Emerg Med, 2005
Mannitol Dose-ResponseMannitol Dose-Response
Sorani J Neurotrauma 2008
Physiology Cluster Analysis
PbtO2
ETCO2
SBPDBPMAP
Self-organizing map reduces high-dimensional information to a two-dimensional grid
Sorani Neurocritical Care 2007
UCSF Next (and Current) EffortsUCSF Next (and Current) Efforts
• Create group identityCreate group identity– C-BICC – Center for Biomedical Informatics in Critical CareC-BICC – Center for Biomedical Informatics in Critical Care
• Obtain fundingObtain funding
• Develop data warehouseDevelop data warehouse
• Undertake advanced informatics and statistical analyses toUndertake advanced informatics and statistical analyses to– Remove artifactsRemove artifacts– Identify event signaturesIdentify event signatures– Improve data visualizationImprove data visualization
• Allow some use for hospital QA Allow some use for hospital QA (helps with administrative buy-in)(helps with administrative buy-in)
• PublishPublish
NeuroICU Physiological InformaticsNeuroICU Physiological Informatics
• Collaborative ProjectCollaborative Project– Admit it: this is beyond bedside cliniciansAdmit it: this is beyond bedside clinicians– Clinicians, computer scientists, informatics, Clinicians, computer scientists, informatics,
industryindustry
• UC Discovery GrantUC Discovery Grant– Pilot project between UCSF, UC Berkeley, IntelPilot project between UCSF, UC Berkeley, Intel– Two years: develop data warehouse methods, Two years: develop data warehouse methods,
pilot data analysispilot data analysis– Expand to multi-center project (will require large Expand to multi-center project (will require large
numbers of patients with long-term outcome)numbers of patients with long-term outcome)
• NIH/NINDS SBIR – Scott Winterstein, PhDNIH/NINDS SBIR – Scott Winterstein, PhD– Data acquisition methodology and device libraryData acquisition methodology and device library
NICU Data Acquisition 2009NICU Data Acquisition 2009
• The primary data are:The primary data are:1.1. Bedside physiological data (Aristein-”homemade”)Bedside physiological data (Aristein-”homemade”)2.2. ICU Patient Care Chart (Carevue-Philips)ICU Patient Care Chart (Carevue-Philips)3.3. Lifetime Clinical Record (Invision-Siemens)Lifetime Clinical Record (Invision-Siemens)
• No kiosk – each bed with networked data acquisitionNo kiosk – each bed with networked data acquisition• Bedside physiological data collected continuously (Q1 minute) and Bedside physiological data collected continuously (Q1 minute) and
automatically into Data Registry Serverautomatically into Data Registry Server
• Must have contextual data (e.g. medications and timing) in order to Must have contextual data (e.g. medications and timing) in order to make sense make sense of physiological dataof physiological data
NICU Data Acquisition 2009NICU Data Acquisition 2009
I SM D atam ar t
A r istein SQ L
I n v ision SQ L
Stor ag e A r eaN etw or kStor ag e
M etadataW ar eh ouse
SF G HF ir ew al l
D ata Sources System Servers
Stag in gSer ver
E T LSer ver
Q B 3Ser ver
D ata T ransm ittedto Q B 3
I nvisio n
SiemensD em ographics
A ristein
A risteinB io informatics
P hysio logy
C areV ue
P hi l ipsN ursing
D ocum entation
P ersonalH ealth
I nform ation
D e- i d en ti fi ed d ata
• Current databaseCurrent database
– CareVue data on CareVue data on ~11,000 patients~11,000 patients
– Physiology data on Physiology data on ~1000 patients~1000 patients
Query Building Screen
Number of patients in data set(current test data)
Data sources and filters
Invision LCR
Aristein high frequency physiology
CareVueNursing documentation of medications, treatments, assessments, laboratory values, IV solutions administered
Once filters have been selected, the user clickson show patients to see preliminary data.
Query Building Screen
Number of patients in data set(current test data)
Data sources and filters
Invision LCR
Aristein high frequency physiology
CareVueNursing documentation of medications, treatments, assessments, laboratory values, IV solutions administered
Once filters have been selected, the user clickson show patients to see preliminary data.
Number of patients meeting selection
Preliminary results show the number of rows of data per variable per patient. The 3 data sources provide 60 possible variables. This screen shot shows only a subset of physiologic variables.
Rows of data for this patient and this variable.
User can download data into a csv file for a single patient or all patients at one time.
Number of patients meeting selection
Preliminary results show the number of rows of data per variable per patient. The 3 data sources provide 60 possible variables. This screen shot shows only a subset of physiologic variables.
Rows of data for this patient and this variable.
User can download data into a csv file for a single patient or all patients at one time.
Sample data. Data displayed in a spreadsheet format.
Shows subset of available variables from 3 data sources
Data are integrated by date/time stamp.
A = physiology
M = medication data
I = Intake or Output data
C = nursing documentation of treatments or assessments
Sample data. Data displayed in a spreadsheet format.
Shows subset of available variables from 3 data sources
Data are integrated by date/time stamp.
A = physiology
M = medication data
I = Intake or Output data
C = nursing documentation of treatments or assessments
Novel Data Visualization ToolsNovel Data Visualization Tools
• Viewing large amounts of data in Viewing large amounts of data in clinically useful wayclinically useful way
• Medications and eventsMedications and events
• Compressed time scalesCompressed time scales
• Physiological “signatures”Physiological “signatures”
Patient Applications: Data Visualization
36 days of continuous physiological data
Acetaminophen then antibiotics
State 1 State 2 State 3 State 4 States 5,6 State 7
PHYSIOLOGIC SIGNATURES
Pattern Recognition
Dynamic Bayesian Networks
We treat patients as if we are practicing DBN state theory.
No really, we do.
Our ProblemsOur Problems
• Paying for all thisPaying for all this– PersonnelPersonnel– Data warehousing (ongoing)Data warehousing (ongoing)– Business models of for-profit companies Business models of for-profit companies
(“just contract with Oracle”) don’t currently (“just contract with Oracle”) don’t currently work for research needswork for research needs
• BalanceBalance– Just like doctors have different specialties, Just like doctors have different specialties,
so do engineers, programmers, so do engineers, programmers, database/informatics experts, statisticians, database/informatics experts, statisticians, computer scientistscomputer scientists
– Clinical coordination – responsible for Clinical coordination – responsible for publishing in clinical journalspublishing in clinical journals
Evidence-based Neurocritical CareEvidence-based Neurocritical Care
• Expertise mattersExpertise matters
• Pronovost, Pronovost, JAMAJAMA, 2002 – systematic review of 26 studies, 2002 – systematic review of 26 studies– Presence of intensivist ass. w/ better outcomesPresence of intensivist ass. w/ better outcomes– Only 1 neuroICU studiedOnly 1 neuroICU studied
• Neurointensivists – improved outcomeNeurointensivists – improved outcome– Suarez, Suarez, Critical Care MedicineCritical Care Medicine, 2004, 2004– Varelas, Varelas, Critical Care MedicineCritical Care Medicine, 2004, 2004
» Semi-closed unit; 30% TBISemi-closed unit; 30% TBI
• UnderstandingUnderstanding– Why expertise makes a difference even without a specific Why expertise makes a difference even without a specific
obvious treatmentobvious treatment– How to harness and “export” expertiseHow to harness and “export” expertise
UCSF ICU Informatics – Guiding PrinciplesUCSF ICU Informatics – Guiding Principles
• NeuroICU monitoring tools have advanced beyond our current NeuroICU monitoring tools have advanced beyond our current ability to understand how to use themability to understand how to use them
• This is due to the disconnect between data This is due to the disconnect between data generationgeneration and and data data analysisanalysis
• Advances in real-time user-friendly data analysis must Advances in real-time user-friendly data analysis must accompany advances in neuromonitoring techniquesaccompany advances in neuromonitoring techniques
• This will be a “long haul”This will be a “long haul”• This is a large-scale collaborative effort across institutionsThis is a large-scale collaborative effort across institutions• Avoid the temptations toAvoid the temptations to
– Be impatient and give upBe impatient and give up– Assume the data we want is easily obtained/acquiredAssume the data we want is easily obtained/acquired– Expect big answers right awayExpect big answers right away– Read too much into early simple analysesRead too much into early simple analyses– Assume large companies will provide us with the solutionsAssume large companies will provide us with the solutions
• Publish – interim experience and results must be disseminatedPublish – interim experience and results must be disseminated
UCSF Neurosurgery
Geoff Manley, MD, PhD
Diane Morabito, RN MPH
Guy Rosenthal, MD
Michele Meeker, RN
Scott Winterstein, PhD
Acknowledgements
UCSF Neuroradiology
Pratik Mukherjee, MD PhD
Alisa Gean, MD
Brain Trauma Foundation
Jam Ghajar, MD PhD
UCSF Neurology
Wade Smith, MD,PhD
UCSF Medical Informatics
Marco Sorani
UC Berkeley Computer Science
Stuart Russell
Norm Aleks
Intel Corporation
Doug Busch
Kevin Conlon
UC Berkeley Neuroscience Institute
Robert Knight, MD
NIH R01NS050173, CDC R49CE000460, NIH K23NS041240 , NIH U10NS058931,
NIH R43NS056639 , UC Discovery Program,
McDonnell-Pew Foundation