gilles clermont: modeling critical illness
TRANSCRIPT
CC∙RR∙II∙SS∙MM∙AACritical Care Medicine
the University of Pittsburgh
Modeling Critical Illness
Center for Inflammation and Regenerative Modeling (CIRM) and The CRISMA Laboratory
Critical Care MedicineSchool of Medicine
University of Pittsburgh
Gilles Clermont MD, MSc
ObjectivesObjectives
The clinical problems facing Critical The clinical problems facing Critical IllnessIllnessOur work at the Center for Regenerative Our work at the Center for Regenerative and Inflammatory Modelingand Inflammatory ModelingThe challenges We faceThe challenges We face
The Goal of Critical Care ?The Goal of Critical Care ?
Health
Disease
DeathZone of opportunity
The big challenges in Critical IllnessThe big challenges in Critical Illness
Timely diagnosisTimely diagnosisOutcome predictionOutcome predictionDevelopment of targetted Development of targetted (personalized) therapies(personalized) therapies
Modulation of the inflammatory responseModulation of the inflammatory response
Root causes of these challengesRoot causes of these challenges
Insufficient, inaccurate dataInsufficient, inaccurate dataWhat to measureWhat to measurePoint-of-care technologiesPoint-of-care technologies
Insufficient interpretative frameworkInsufficient interpretative frameworkUncertain biological mechanismsUncertain biological mechanismsBiological variabilityBiological variabilityDearth of Dearth of in silicoin silico disease models disease models
Insufficient mathematicsInsufficient mathematics
Critical IllnessesCritical Illnesses
Inflammation
Severe infections(sepsis)
Trauma/Shock
Acute coronary syndrome
Stroke
Anti-InflammationInflam-Inflam-
mationmation
OrganInjury
RecoveryRecovery
TimeTime
INS
UL
TIN
SU
LT
Current Paradigm of Injury/RecoveryCurrent Paradigm of Injury/Recovery
•Innate immunity•Coagulation•Metabolism
Stress responseStress response
The canvas…of opportunititesThe canvas…of opportunitites
Infection•Pathogen•Toxins
Early mediators•TNF•IL-1•IL-10•IL-6•Cells
Detection•Dendritic cells•Macrophages
Coagulation
Physiologicmanifestations
Metabolicmanifestations
Organdysfunction
Late mediators
Death
Opportunities for Immune SupportOpportunities for Immune Support
Time
IL-10 RECOVERY
INSULT
IL-6TNF
IL-1
Anti-LPSAnti-TNFAnti-IL-1Anti-IL-10
Blood Purification
Immunologic Support
HMGB1
Anti-HMGB1
What about immunomodulation in What about immunomodulation in sepsis and trauma?sepsis and trauma?
25 years of global disaster25 years of global disasterSimplistic rationalesSimplistic rationalesIneffective productsIneffective productsPoor patient selectionPoor patient selection
>70 phase II trials, > 1B dollars invested>70 phase II trials, > 1B dollars investedTwo possible leadsTwo possible leads
Low dose anti-inflammatory treatmentLow dose anti-inflammatory treatmentActivated protein CActivated protein C
Contradictory results!Contradictory results!
Treating sepsis: the strength of the Treating sepsis: the strength of the ConsensusConsensus
2532
18
Good or Bad?
The inflammatory responseThe inflammatory response
Huang Q, Science 2001
Calvano, Nature 2005
Modeling Inflammation at the CIRMModeling Inflammation at the CIRM
Multiple models of acute inflammation (sepsis, trauma/hemorrhage, Multiple models of acute inflammation (sepsis, trauma/hemorrhage, biowarfare agents, phonotrauma, wound healing), organ biowarfare agents, phonotrauma, wound healing), organ damage/dysfunction, and healing/regenerationdamage/dysfunction, and healing/regeneration
Qualitative and quantitative predictionsQualitative and quantitative predictionsProbing mechanismsProbing mechanisms
Have simulated device usage and guided device designHave simulated device usage and guided device designHave outlined an iterative strategy for rational drug design and Have outlined an iterative strategy for rational drug design and administrationadministrationHave carried out simulated clinical trials in the settings of sepsis Have carried out simulated clinical trials in the settings of sepsis and trauma, including biowarfare applicationsand trauma, including biowarfare applications
Small models of inflammationSmall models of inflammation
Top-downTop-down““Understand” the biologyUnderstand” the biology
Biological plausibilityBiological plausibility
High-level “map” of the biologyHigh-level “map” of the biologyBuilding blocs for more complex models Building blocs for more complex models
Reduced models of inflammationReduced models of inflammation
Transients for 3 possible regimenTransients for 3 possible regimenHealth
Septic death
Aseptic death
Bifurcation analysis on KpgBifurcation analysis on Kpg
“Septic death”
“Aseptic death”
“Heatlh”
2-D bifurcation diagram2-D bifurcation diagram
Opportunity
Manipulating anti-inflammatoriesManipulating anti-inflammatories
Why complicate thingsWhy complicate things
To produce a calibratedTo produce a calibratedTo “intervene” in the dynamics in a To “intervene” in the dynamics in a realistic way, more realistic “handles” realistic way, more realistic “handles” are neededare neededNot all “modules” need to be equally Not all “modules” need to be equally detaileddetailedThe analysis of large models:The analysis of large models:
May rapidly become intractable May rapidly become intractable May not yield useful resultsMay not yield useful results
Simulating Inflammatory Disorders at the Simulating Inflammatory Disorders at the CIRMCIRM
ResearchBiological
Mechanisms
DevelopRepresentative
Models
Collect Biomarker
Data
CalibrateModels to Data
Use Modelfor
Predictions
A Unified Inflammatory ResponseA Unified Inflammatory Response
Simulations of infectious agents with Simulations of infectious agents with bioterror potentialbioterror potential
Shock 2007
JTB 2007
Probing mechanismsProbing mechanisms
In silicoIn silico design of RCTs design of RCTs
Anti-TNF treatment for sepsis:Anti-TNF treatment for sepsis:A simulation studyA simulation study
Clermont et al. 2004
Pat
ho
gen
vir
ule
nce
(Qu
arti
le)
Pat
ho
gen
lo
ad(Q
uar
tile
)
Q1
Q2
Q3
Q4
0% 25% 50% 75% 100%Percent
TN
F r
es
po
ns
ive
ne
ss
(Qu
art
ile
)
Q1
Q2
Q3
Q4
An
ti-i
nfl
amm
ato
ry r
esp
on
sive
nes
s(Q
uar
tile
)Q1
Q2
Q3
Q4
0% 25% 50% 75% 100%Percent
Q1
Q2
Q3
Q4
Helped by treatment Lived irrespective of treatmentDied irrespective of treatment Harmed by treatment
Outcome by subgroupsOutcome by subgroups
Validating Validating in silicoin silico simulation? simulation?
Face validity of the disease modelFace validity of the disease modelKnowledge of key driving factorsKnowledge of key driving factorsDisease model includes these factorsDisease model includes these factorsIntervention modelIntervention model
Mechanism of action of the proposed treatmentMechanism of action of the proposed treatmentPK/PD dataPK/PD data
Predictive ability on empirical dataPredictive ability on empirical dataControlled experimentsControlled experimentsExisting trial dataExisting trial data
Disease modelDisease model
Must account for uncertain mechanismsMust account for uncertain mechanismsModel structure recapitulates biologyModel structure recapitulates biology
Predictors in a statistical modelPredictors in a statistical modelEquations/rules in white box modelsEquations/rules in white box models
Must make best use of observations at Must make best use of observations at hand which are often incompletehand which are often incomplete
Within a given model structure, develop an Within a given model structure, develop an understanding of the breath of parameter understanding of the breath of parameter realizations that fit data equally wellrealizations that fit data equally well
Uncertainty in the relative importance of Uncertainty in the relative importance of mechanisms/interactionsmechanisms/interactions
Many model realizations are necessaryMany model realizations are necessary
Some core theoretical challengesSome core theoretical challenges
The variability problemThe variability problemThe inverse problemThe inverse problemThe “cogent-reduction” problemThe “cogent-reduction” problem
The variability problemThe variability problem
Can inter-individual variability be Can inter-individual variability be characterized in some fashion?characterized in some fashion?
Sepsis as an exampleSepsis as an example
How can mathematical models to How can mathematical models to capture variability?capture variability?Can meaningful diagnostic and Can meaningful diagnostic and therapeutic insights be achieved in the therapeutic insights be achieved in the presence of such variability?presence of such variability?
Day 1 cytokine levels in non-septic Day 1 cytokine levels in non-septic patients for prediction of severe sepsispatients for prediction of severe sepsis
TN
F a
t ba
selin
e a
nd 9
5% C
I (ln
pg/
ml)
2.1
2.0
1.9
1.8
1.7
p = 0.0087
Severe sepsis(n=268)
No SS(n=1073)
TNF
Analysis restricted to day 1 levels of those patients who do NOT have severe sepsis on first day
4.0
3.8
3.6
3.4
3.2IL6
at b
ase
line
and
95%
CI
p = 0.0009
Severe sepsis(n=268)
No SS(n=1073)
IL-6
Day 1 levels and survivalDay 1 levels and survival
Dead(n=212)
Alive(n=1410)
Ba
selin
e T
NF
with
95
% C
I
2.4
2.2
2.0
1.8
p<0.0001
5.0
4.6
4.2
3.8
3.4Dead(n=212)
Alive(n=1410)
Ba
selin
e I
L6
with
95
% C
I
p<0.0001
2.8
2.6
2.4
2.2
Dead(n=212)
Alive(n=1410)
Ba
selin
e I
L1
0 w
ith 9
5%
CI
p<0.0001
TNF
IL-10
IL-6
Day 1 cytokine levels in patients who Day 1 cytokine levels in patients who develop ARF and those that do notdevelop ARF and those that do not
ARF = RIFLE-I or F
4
4.5
5
5.5
6
6.5
7
7.5TNF
Pg
/ml
SE
M
ARF(n=258)
No ARF(n=1544)
p<0.0001
IL-6P
g/m
l S
EM
ARF(n=258)
No ARF(n=1544)
p<0.0001
20
30
40
50
60
70
80IL-10
Pg
/ml
SE
M
ARF(n=258)
No ARF(n=1544)
p=0.0285
4
4.5
5
5.5
6
6.5
7
7.5
TrajectoriesTrajectories
Trajectories are consistent within individualsTrajectories are consistent within individuals““Shapes” are often consistent across Shapes” are often consistent across individualsindividuals
LPS 4 ng/kg IV
0
1000
2000
3000
4000
5000
6000
0 1 2 3 4 5 6
Time (h)
#1
#7
#9
#21
LPS (1 ng/kg IV)
0
50
100
150
200
250
300
0 1 2 3 4 5 6
Time (h)
seru
m T
NF
(p
g/m
l)
#2
#5
#13
#14
Suffredini et al. 1996
Cytokines by Outcome (60 days)Cytokines by Outcome (60 days)
IL-6
0
20
40
60
80
100
120
140
1 2 3 4 5 6 7
Days
pg
/ml
SSD
SSA
NSS
IL-10
0
2
4
6
8
10
12
1 2 3 4 5 6 7
Days
pg
/ml
SSD
SSA
NSS
TNF
0
2
4
6
8
10
1 2 3 4 5 6 7
Days
pg
/ml
SSD
SSA
NSS
Trajectory AnalysisTrajectory Analysis
1 2 3 4 5 6 7
Day
Log IL-6
Day
Log IL-10
1 2 3 4 5 6 7
HH
MM
LL
Some good newsSome good news
Standard (although by no means Standard (although by no means elementary) statistical techniques elementary) statistical techniques identify “classes” of patientsidentify “classes” of patients
PhysiologyPhysiologyOmicsOmics
Qualitative patterns, but not magnitude Qualitative patterns, but not magnitude of response, often preserved across of response, often preserved across individualsindividuals
Within speciesWithin species
One model – one patientOne model – one patient
M
M
M
Disease modelsDisease models
Must account for uncertain mechanismsMust account for uncertain mechanismsModel structure recapitulates biologyModel structure recapitulates biology
Predictors in a statistical modelPredictors in a statistical modelEquations/rules in white box modelsEquations/rules in white box models
Must make best use of observations at Must make best use of observations at hand which are often incompletehand which are often incomplete
Within a given model structure, develop an Within a given model structure, develop an understanding of the breath of parameter understanding of the breath of parameter realizations that fit data equally wellrealizations that fit data equally well
Uncertainty in the relative importance of Uncertainty in the relative importance of mechanisms/interactionsmechanisms/interactions
Many model realizations are necessaryMany model realizations are necessary
Patient-specific metamodelPatient-specific metamodel
M1
M2
Mn
E(Mn) ≡ Metamodel or Ensemble
Where the individual models vary in theirmathematical structure and parameters
Population-level EnsemblePopulation-level Ensemble
E(Mn)E(Mp)
E(Mq)
• Empirical rather than phenomenological
Targetted TherapyTargetted TherapyModel Predictive Control (MPC)Model Predictive Control (MPC)
• The desired output is “health” • The MPC method uses actual data and model simulations
to estimate output: the discrepancy is estimated (Sensor)• The MPC method suggest an optimal intention strategy
which is time dependent (Actuator)
“Base” System• Real patient• MetamodelINPUT
•Actual dataOUTPUT•Actual data•Predicted data
Sensor• Error between actual/desired
Actuator
Controller
Horizon
k k+1 k+2 k+m-1 k+h
Past Future
R(k-2)
p(k-2)
M(k-2)
R(k)
p(k)
M(k)
R(k+1)R(k+2)
u(k+m-1)
uk
p(k+1)
Reference Trajectory, R
PredictedOutput, p
Measured Output, M
Control Action, u
MPC Schematic
ΔU
Babatunde A. Ogunnaike and W. Harmon Ray. Process Dynamics, Modeling, and Control (Topics in Chemical Engineering). Oxford University Press: New York, 1994. pg. 997.
m – move horizonh – prediction horizonk – current simulation time step
Patient 20
Patient 405
TailoredStandard
Early treatment, frequent titration are key Early treatment, frequent titration are key Measurement frequency congruent with the time Measurement frequency congruent with the time scale of the process being modulated is keyscale of the process being modulated is key
The inverse problemThe inverse problem
time
Observe Low Bloodpressure
Observe Bloodpressure
Give afluidchallenge
Low
Normal
Mid-range
Model reduction – Bottom-up modelsModel reduction – Bottom-up models
Map “lumped observables” to Map “lumped observables” to biologically relevant functions/modulesbiologically relevant functions/modules
Component aggregationComponent aggregation Identify aggregates (SVD – PCA)Identify aggregates (SVD – PCA)
Time aggregationTime aggregationSpatial segregationSpatial segregation
Data-driven vs. Knowledge-drivenData-driven vs. Knowledge-driven
Model reduction – Top-down modelsModel reduction – Top-down models
Function vs components may be biasedFunction vs components may be biasedWhat we think we knowWhat we think we knowWhat we can measureWhat we can measure
Which parts of the model can/should be Which parts of the model can/should be abstracted abstracted Could this be driven by constraints Could this be driven by constraints imposed by biological laws?imposed by biological laws?
Knowledge and successful translationKnowledge and successful translation
EmpiricalEpicycles
- to XVII century
Kepler/Newton Ellipses
First interactionbetween a physicallaw and empiricobservation
GR - EinsteinPrecessing ellipses
Discrepancy betweenpredictions and empiric observation
Depth of knowledge
Discrepancy betweengravity and otherforces of nature
QG - ??Black hole physics
Inflammation Modeling is a Team SportInflammation Modeling is a Team Sport
Mathematics (Pitt)Carson ChowBard ErmentroutJonathan RubinBeatrice RiviereIvan YotovDavid SwigonJudy DayQi Mi
Mathematics (CMU)Shlomo Ta’asanRima Gandlin
Statistics (Pitt)Greg Constantine
Immunetrics, Inc.John BartelsSteve ChangArie BarattJoyce Wei
IBMFred Busche
Cook County HospitalGary An
University of CologneEddy NeugebauerRolf Lefering
Ludwig Boltzmann InstituteHeinz Redl
SUNY-UpstateGary NiemanDavid Carney
Surgery (Pitt)Tim BilliarRuben ZamoraRosie HoffmanDavid HackamRobert KormosDavid SteedEdith TzengJuan OchoaClaudio LagoaAndres TorresBinnie BittenDerek BarclayThierry Clermont
Critical Care Medicine (Pitt)Gilles ClermontMitchell FinkJohn KellumRuss DeludeJuan Carlos Puyana
McGowan Institute (Pitt)Alan RussellJohn MurphyWilliam FederspielWilliam Wagner
SHRS (Pitt)Cliff BrubakerKittie Verdolini
Medicine (Pitt)David WhitcombMarc Roberts
Children’s Hospital of PittsburghDavid HackamJeffrey UppermanPat HebdaRaphael Hirsch
www.iccai.org
CC∙RR∙II∙SS∙MM∙AACritical Care Medicine
the University of PittsburghInflammation Modeling is a Team Sport
Mathematics (Pitt)Mathematics (Pitt)Carson ChowCarson ChowBard ErmentroutBard ErmentroutJonathan RubinJonathan RubinBeatrice RiviereBeatrice RiviereIvan YotovIvan YotovDavid SwigonDavid SwigonJudy DayJudy DayQi MiQi Mi
Mathematics (CMU)Mathematics (CMU)Shlomo Ta’asanShlomo Ta’asanRima GandlinRima Gandlin
Statistics (Pitt)Statistics (Pitt)Greg ConstantineGreg Constantine
Immunetrics, Inc.Immunetrics, Inc.John BartelsJohn BartelsSteve ChangSteve ChangArie BarattArie BarattJoyce WeiJoyce Wei
IBMIBMFred BuscheFred Busche
Cook County HospitalCook County HospitalGary AnGary An
University of CologneUniversity of CologneEddy NeugebauerEddy NeugebauerRolf LeferingRolf Lefering
Ludwig Boltzmann InstituteLudwig Boltzmann InstituteHeinz RedlHeinz Redl
SUNY-UpstateSUNY-UpstateGary NiemanGary NiemanDavid CarneyDavid Carney
Surgery (Pitt)Surgery (Pitt)Tim BilliarTim BilliarRuben ZamoraRuben ZamoraRosie HoffmanRosie HoffmanDavid HackamDavid HackamRobert KormosRobert KormosDavid SteedDavid SteedEdith TzengEdith TzengJuan OchoaJuan OchoaClaudio LagoaClaudio LagoaAndres TorresAndres TorresBinnie BittenBinnie BittenDerek BarclayDerek BarclayThierry ClermontThierry Clermont
Critical Care Medicine (Pitt)Critical Care Medicine (Pitt)Gilles ClermontGilles ClermontMitchell FinkMitchell FinkJohn KellumJohn KellumRuss DeludeRuss DeludeJuan Carlos PuyanaJuan Carlos Puyana
McGowan Institute (Pitt)McGowan Institute (Pitt)Alan RussellAlan RussellJohn MurphyJohn MurphyWilliam FederspielWilliam FederspielWilliam WagnerWilliam Wagner
SHRS (Pitt)SHRS (Pitt)Cliff BrubakerCliff BrubakerKittie VerdoliniKittie Verdolini
Medicine (Pitt)Medicine (Pitt)David WhitcombDavid WhitcombMarc RobertsMarc Roberts
Children’s Hospital of PittsburghChildren’s Hospital of PittsburghDavid HackamDavid HackamJeffrey UppermanJeffrey UppermanPat HebdaPat HebdaRaphael HirschRaphael Hirsch
Developing Trans-disciplinary Mathematical Modeling Teams for the study of Acute Inflammation: A review of the Experiences of Four Research Groups Gary An MD (1), C. Antony Hunt PhD (2), Gilles Clermont MD (3, 6), Edmund Neugebauer, PhD (4) and Yoram Vodovotz PhD (5, 6)
1) Department of Surgery, Northwestern University Feinberg School of Medicine 2) Biosystems Research Group, University of California, San Francisco 3) Department of Critical Care Medicine, University of Pittsburgh4) Institute for Research in Operative Medicine, University of Witten/Herdecke5) Departments of Surgery and Immunology, University of Pittsburgh 6) Center for Inflammation and Regenerative Modeling, McGowan Institute for Regenerative Medicine, University of Pittsburgh, Pittsburgh, PA
J. Crit. Care, submitted for publication
Developing Trans-disciplinary Mathematical Modeling Teams for the study of Acute Inflammation: A review of the Experiences of Four Research Groups Gary An MD (1), C. Antony Hunt PhD (2), Gilles Clermont MD (3, 6), Edmund Neugebauer, PhD (4) and Yoram Vodovotz PhD (5, 6)
1) Department of Surgery, Northwestern University Feinberg School of Medicine 2) Biosystems Research Group, University of California, San Francisco 3) Department of Critical Care Medicine, University of Pittsburgh4) Institute for Research in Operative Medicine, University of Witten/Herdecke5) Departments of Surgery and Immunology, University of Pittsburgh 6) Center for Inflammation and Regenerative Modeling, McGowan Institute for Regenerative Medicine, University of Pittsburgh, Pittsburgh, PA
J. Crit. Care, submitted for publication
International Conference on Complexity in Acute Illness
http://www.scai-med.org