Transcript Estimation of Subject Specific ICP Dynamic Models Using

Estimation of Subject Specific ICP Dynamic
Models Using Prospective Clinical Data
Biomedicine 2005, Bologna, Italy
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
W. Wakeland
1,2,
J. Fusion 1, B. Goldstein
3
1
Systems Science Ph.D. Program, Portland State University, Portland, Oregon, USA
2 Biomedical Signal Processing Laboratory, Department of Electrical and Computer
Engineering, Portland State University, Portland, Oregon, USA
3 Complex Systems Laboratory, Doernbecher Children’s Hospital, Division of Pediatric
Critical Care, Oregon Health & Science University, Portland, Oregon, USA
This work was supported in part by the Thrasher Research Fund
E LECTRICAL & COMPUTER
E NGINEERING
System Science
Ph.D. Program
Oregon Health & Science Univ.
1
Complex Systems Laboratory
Aim
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• To develop tools for improving care of children
with severe traumatic brain injury (TBI)
Help improve diagnosis and treatment of
elevated intracranial pressure (ICP)
Improve long-term outcome following severe
TBI
• One potential approach:
Create subject-specific computer models of ICP
dynamics
Use models to evaluate therapeutic options
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
2
Motivation
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• TBI is the leading cause of death and disability
in children
150,000 pediatric brain injuries
7,000 deaths annually (50% of all childhood
deaths)
29,000 children with new, permanent disabilities
• Death rate for severe TBI (defined as a Glasgow
Coma Scale score < 8) remains between 30%45% at major children's hospitals
• A recently published evidence-based medicine
review reports that elevated ICP is a primary
determinant of outcome following TBI
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
3
Background: Intracranial Pressure (ICP)
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• TBI often causes ICP to increase
Frequently due, at least initially, to internal
bleeding (hematoma)
Elevated ICP is defined as > 20 mmHg
• Persistent elevated ICP  reduced blood flow
 insufficient tissue perfusion (ischemia)
 secondary injury  poor outcome
• Poor outcomes often occur despite the
availability of many treatment options
The pathophysiology is complex and only
partially understood
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
4
Background: Treatment Options
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• Treatment options include, among many
others:
Draining cerebral spinal fluid (CSF) via a
ventriculostomy catheter
Raising the head-of-bed (HOB) elevation to 30
to promote jugular venous drainage
Inducing mild hyperventilation
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
5
Background: ICP Dynamic Modeling
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• Many computer models of ICP have been
developed over the past 30 years
Models have sophisticated logic (differential eqns.)
Potentially very helpful in a clinical setting
• However, clinical impact of models has been
minimal
Complex models are difficult to understand and use
• Another issue is that clinical data often lack the
annotations needed to facilitate modeling
Exact timing for medications, CSF drainage,
ventilator adjustments, etc.
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
6
Method: Research Approach
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• Use an experiment protocol (next slide) to collect
prospective clinical data
Physiologic signals recorded continuously
electrocardiogram, respiration, arterial blood
pressure, ICP, oxygen saturation
Plus annotations to indicate the precise timing of
therapies and physiologic challenges
• Use collected data to create subject-specific
computer models of ICP dynamics
• Use subject-specific models to predict patient
response to treatment and challenges
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
7
Method: Experimental Protocol
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• Mild physiologic challenges
Applied over multiple iterations to three subjects
with severe traumatic brain injury
• Change the angle of the head of the bed (HOB)
Randomly assigned, between 0º and 40º, in 10º
increments, for 10 minute intervals
• Change minute ventilation (or respiration rate,
RR)
Clinician adjusts RR to achieve specified ETCO2
target from [-3 to -4] mmHg to [+3 to +4] mmHg
from baseline
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
8
Method: Model Estimation
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
Initial
Parameters
Nonlinear
Optimizing
Algorithm
HOB and RR
Challenges
Estimated
Parameters
ICP
Dynamic
Model
Error
Predicted ICP
Error
Computation
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Measured ICP
Oregon Health & Science Univ.
Complex Systems Laboratory
9
Method: Simulink ICP Dynamic Model
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
10
Method: Model, Core Logic
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• The timing for physiologic challenges is a key
input to the model
• The state variables are the volumes of each
fluid compartment
• Key feedback loops
Volume pressure  flow  volume
∑ (volumes)  ICP  pressures  flows 
∑ (volumes)
• Autoregulation is modeled by changing arterialto-capillary flow resistance [only]
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
11
Method: Model, Impact of Challenges
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• Impact of HOB angle (ө) on ICP
↑ө
intracranial arterial pressure ↓
intracranial venous pressure ↓
ICP↓
• Impact of RR on ICP
↑RR
PaCO2 ↓
indicated blood flow ↓
ICP↓
capillary resistance ↑
arterial blood volume ↓
arterial-to-capillary flow ↓
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
12
Method: Parameters Estimated
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
•
•
•
•
•
•
•
•
Autoregulation factor
Basal cranial volume
CSF drainage rate
Hematoma increase rate
 pressure time constant
ETCO2 time constant
Smooth muscle “gain constant”
Systemic venous pressure
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
13
Results: Patient 1, Session 4. A series of
changes
to
HOB
elevation
and
RR
B
S
P
L
IOMEDICAL
IGNAL
ROCESSING
ABORATORY
bsp.pdx.edu
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
14
Results: Patient 2, Session 1. A series of
changes
to
HOB
elevation
S
P
L
BIOMEDICAL
IGNAL
ROCESSING
ABORATORY
bsp.pdx.edu
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
15
Results: Patient 2, Session 4. A series of
changes to RR
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
16
Results: Patient 2, Session 7. A series of
changes
to
HOB
elevation
and
RR
S
P
L
BIOMEDICAL
IGNAL
ROCESSING
ABORATORY
bsp.pdx.edu
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
17
Results: Summary
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
18
Discussion: Model vs. Actual Response
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• Model response to HOB changes was very
similar to actual response (error < 1 mmHg)
• Response to RR changes did not fully reflect
the patient’s actual response in all cases
Error > 2 mmHg in many cases
Revealed several model deficiencies
Lack of systemic adaptation
Does not capture interaction affects
Incorrect response to RR changes
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
19
Discussion: Model Deficiencies
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• Systemic adaptation (make change; return to baseline)
 P2S7: When HOB moved from 30º to 0º; then back to 30º,
the ending in vivo ICP was lower than its starting point
 In the model, ICP returned to its original value
• Interaction of interventions
 ICP impact depended on whether the interventions were
temporally clustered or dispersed
 Model did not capture these differences
• Incorrect model response to RR changes
 Changes in smooth muscle tone in the model affect the
arterial-to-capillary blood flow resistance, but not
[directly] the arterial volume
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
20
Discussion: Summary
BIOMEDICAL SIGNAL PROCESSING LABORATORY
bsp.pdx.edu
• Model of ICP dynamics was calibrated to
replicate the ICP recorded from specifics patient
during an experimental protocol
• Results demonstrated the potential for using
clinically annotated prospective data to create
subject-specific computer simulation models
• Future research will focus on improving the logic
for cerebral autoregulatory mechanisms and
physiologic adaptation
E LECTRICAL & COMPUTER System Science
E NGINEERING
Ph.D. Program
Oregon Health & Science Univ.
Complex Systems Laboratory
21