PREDICTING TIME TO DEATH IN DCD & THE IMPACT OF WLSM ... and Dying T… · Happened within 30...
Transcript of PREDICTING TIME TO DEATH IN DCD & THE IMPACT OF WLSM ... and Dying T… · Happened within 30...
DEATH….
THE GREAT UNKNOWN
PREDICTING TIME TO DEATH IN DCD & THE IMPACT OF WLSM PROCEDURES
Dr Jason Shahin
MDCM FRCPC MSC
McGill University
Dept of Critical Care
TWO QUESTIONS TODAY?
1) When do we die
2) How do we die
QUESTION 1- WHEN DO WE DIEDCD-BARRIERS AND CONTROVERSIES
• Time period required to confirm “irreversible death”
• Autoresuscitation
• Ethical considerations “violation of the dead donor rule”
• Time to death prediction
- warm ischemia
DCD OVERVIEW60-120 minutes
DCD OVERVIEW60-120 minutes
Withdrawal of life support
therapy
Systematic review
DEPPART SUB-STUDY OVERVIEW
Research Question:
Can time to death be predicted ?
Primary Objective:
To develop a model that can predict the time from WLSM to death in
potential DCD donors
7
STUDY DESIGN & GOALS
Prospective observational study
306 patients from 19 ICUs in Canada (12 adult, 2
pediatric), Czech republic and The Netherlands
Inclusion criteria: all DCD eligible patients where
imminent death anticipated and a decision for WLSM
had been made
8
`
WLSM
•Vitals/Neuro exam
•First act of WLSM
•Time of extubation
•Death Prediction (Physician)
Circulatory, Respiratory, pharmacological 1h before WLSM until declaration of death
Declaration of death
Death
•Labs, CT scan
•APACHE IIDecision to WLSM
+/-Consent to DCD
•Demographic data
•Diagnosis
•Comorbidities
•Length of ICU stay
Consent to study
DATA COLLECTION– TIMELINE
WHAT WE DID
Developed a risk prediction tool
using 2 philosophically different
approaches
1. Classical regression- multivariable
analysis logistic regression
2. Machine learning- Purely
statistical using a Random Forrest
ensemble method
1. Predictors chosen either a priori or based on statistical
significance in sample population
2. Logistic regression used with death as the binary
outcome
3. Predictors may be eliminated based on statistical
association with the outcome to produce a parsimonious
model
CLASSICAL APPROACH
OVERFITTING
OVERFITTING
OVERFITTING
1. Predictors chosen either a priori or based on statistical
significance in sample population
2. Logistic regression used with death as the binary
outcome
3. Predictors may be eliminated based on statistical
association with the outcome to produce a parsimonious
model
CLASSICAL APPROACH
WHAT WE DID
Developed a risk prediction tool
using 2 philosophically different
approaches
1. Classical regression- multivariable
analysis
2. Machine learning- Purely
statistical using a Random Forrest
ensemble method
WHAT IS RANDOM FORREST?
HOW DOES A RANDOM FORREST WORK
HOW DOES A RANDOM FORREST WORK
ANALYSIS AND RESULTS
Expert Opinion
Classical model
13
Machine learning
model
22
Pool of 22
predictors
Predictors from
literature and clinical
judgement
Whole cohort of WLST
639
DCD eligible
306 (47.9%)
Proceeded to DCD
87 (67.9%)
No Organ procurement
21 (24.2%)
Organ procurement
67 (75.8%)
-Time to death deemed too long
-Family uninterested
-SDM refused
Other
Prediction Model
Survival curves for whole cohort
60% of patients died
within 2-hrs
2
TIME TO DEATH-BY 30 MIN INTERVALS
Classical model Random Forest
Full data Accuracy 0.77 0.90
False Negative Rate 0.21 0.10
False Positive Rate 0.27 0.11
Misclassification Rate 0.23 0.10
Area Under the Curve 0.86 0.96
Performance measures for DCD eligible group
Classical
model
Random
Forest
Area Under the Curve 0.86 0.96
Performance measures for DCD eligible group
10-fold cross validationClassical
model
Random
Forest
Area Under the Curve 0.74 0.77
Performance measures for DCD eligible group-validation
Performance measures of prediction models for entire cohort (639)
Classical
model
Random
Forest
Area Under the Curve 0.84 0.92
Performance measures for entire cohort
10-fold cross validation Classical
model
Random
Forest
Area Under the Curve 0.80 0.84
Potential predictors Odds Ratio (95% CI)
Admission Diagnosis (ref: Traumatic brain injury)
Non-traumatic brain injury neurological 0.9 (0.38, 2.14)
Surgery (non-traumatic brain injury) 1.09 (0.26, 4.58)
Medical 1.01 (0.42, 2.46)
Comorbidities (ref: Cardio-respiratory)
Other 1.49 (0.74, 3.00)
None 2.11 (1.01, 4.39) *
BMI (kg/m2) [≥30 vs <30] 1.78 (0.95, 3.33)
APACHE II Score [ref: Score <15]
Score 15-24 0.53 (0.15, 1.81)
Score ≤25 0.06 (0.17, 2.08)
At one-hour pre-WLST
Cardiac arrest with resuscitation in 24 hours pre-WLST [Yes vs No] 0.98 (0.40, 2.38)
Vasopressor use 1-hour pre-WLST [Yes vs No] 0.98 (0.52 1.86)
Systolic blood pressure (mm Hg) [>100 vs ≤ 100] 0.64 (0.32, 0.28)
Opioid analgesic use 1-hour pre-WLST [Yes vs No] 1.65 (0.80, 3.43)
GCS score [3 vs >3] 2.37 (1.26, 4.45) *
Pupillary reflex [Yes vs No] 0.63 (0.2, 1.27)
PaO2 to FiO2 ratio [ref: ≤100]
PaO2 to FiO2 ratio 101-200 0.98 (0.28, 3.40)
PaO2 to FiO2 ratio >200 0.43 (0.13, 1.38)
Respiratory rate (breaths/ minute) [ref: <12]
Respiratory rate 12-25 0.63 (0.23, 1.71)
Respiratory rate >25 0.77 (0.24, 2.47)
Spontaneous respiration [Yes vs No] 0.93 (0.32, 2.71)
Physician’s Prediction [Yes vs No] 7.21 (3.89, 13.38) *
Potential predictors Odds Ratio (95%
CI)
GCS score [3 vs >3] 2.37 (1.26, 4.45) *
Physician’s Prediction [Yes v No] 7.21 (3.89, 13.38) *
• Access to real time data
• Patient trajectory
Predicted outcome
Physician prediction
Physiology
Demographics
&
Co-morbidities
WLSM
method
Predicted outcome
Physician prediction
Physiology
Demographics
&
Co-morbidities
WLSM
method
HOW DO PEOPLE WITHDRAW LIFE SUPPORT?
IS WLSM DIFFERENT FOR DCD CANDIDATES?
QUESTION 2: HOW DO PEOPLE DIE
“The management of the dying
process...should proceed according to
existing ICU practice...not be influenced
by donation potential”
“Thou shalt not do anything different”
WLSM APPROACH
WLSM APPROACH
Reduce ventilator supportReduce ventilator and vasopressors
Extubation only Stopped vasopressors only
Stopped vasopressors and extubated
Multi-step ventilator reduction
WLSM APPROACH
Extubation Stopped vasopressors
Everything else is considered weaning
Vasopressor used
Stopped within 30
minutes of WLSM
Extubated
Happened within 30
minutes of WLSM
Description of Predictor: Vasopressor stopped & Extubated within 30 minutes of WLSM
Considered
vasopressor:
dobutamine
dopamine
ECMO epinephrine
intra-aortic balloon
milrinone
norepinephrine
phenylephrine
terlipressin
vasopressin
Whole cohort of WLSM
639
DCD eligible
205 (32.1%)
Approached for DCD 138 (67.3%)
Proceeded to attempted DC
89 (64.5%)
No Organ procurement
23 (25.9%)
Organ procurement
66 (74.1%)
SDM refused
Other
-DCD not attempted
116 (56.6%)
Relationship between methods of WLSM and time to death [DCD eligible: 205 patients]
Time to death Analysis Crude Analysis Odds ratio lower CI upper CI P-value
30 minAdjusted Vasopressor method[stopped] 1.37 0.87 2.16 0.18
Extubation [Yes] 2.51 0.97 6.50 0.06
1 hourAdjusted Vasopressor method[stopped] 1.44 0.83 2.50 0.20
Extubation [Yes] 2.54 1.15 5.60 0.02
2 hourAdjusted Vasopressor method [stopped] 1.45 0.64 3.27 0.37
Extubation [Yes] 2.16 1.00 4.64 0.05
• Adjusted for age, APACHE II , admission diagnosis and center
clustering
Relationship between methods of WLSM and time to death [DCD eligible: 205 patients]
Time to death Analysis Crude Analysis Odds ratio lower CI upper CI P-value
30 minAdjusted Vasopressor method[stopped] 1.37 0.87 2.16 0.18
Extubation [Yes] 2.51 0.97 6.50 0.06
1 hourAdjusted Vasopressor method[stopped] 1.44 0.83 2.50 0.20
Extubation [Yes] 2.54 1.15 5.60 0.02
2 hourAdjusted Vasopressor method [stopped] 1.45 0.64 3.27 0.37
Extubation [Yes] 2.16 1.00 4.64 0.05
• Adjusted for age, APACHE II , admission diagnosis and center
clustering
DCD Attempted DCD not-attempted p-value
Variables Categories count % count %
89 116
Vasopressor method of WLSM stopped 24 92.3 49 98.0 0.4
Kept on 1 1.14 1 0.86
wean 0 0 0 0
not on-
vasopressor63 71.59 66 56.9
Respiratory method of WLSM Extubated 74 83.15 75 64.66<0.001
not-extubated 5 5.62 38 32.76
wean 10 11.24 3 2.59
Table 1: Methods of WLSM in DCD attempted versus not attempted
DCD Attempted DCD not-attemptedp-value
Variables Categories count % count %
Respiratory
method of WLSMExtubated 74 83.15 75 64.66
<0.001
not-extubated 5 5.62 38 32.76
wean 10 11.24 3 2.59
Table 1: Methods of WLSM in DCD attempted versus not attempted
Table 3: Opioid dose type and total doses [DCD eligible 204 patients]
Opioid delivery of morphine equivalency
Dose type Total dose
count % Mean SD
Bolus only-(mg) 17 9.50 16.65 24.89
Drip only-(mg/hr) 122 68.16 5.97 7.77
Both-(mg) 40 22.35 38 37.38
Table 4: Opioid dose type and total doses by DCD attempt
DCD
attempted
DCD not
attempted
Mean SD Mean SD
Bolus only-mg 21.56 34.55 12.28 12.19
Drip only-mg/hr 4.48 4.86 6.97 9.11
Both-mg 40.64 39.89 33.59 33.64
Table 4: Opioid dose type and total doses by DCD attempt
DCD
attempted
DCD not
attempted
Mean SD Mean SD
Bolus only-mg 21.56 34.55 12.28 12.19
Drip only-mg/hr 4.48 4.86 6.97 9.11
Both-mg 40.64 39.89 33.59 33.64
Table 4: Opioid dose type and total doses by DCD attempt
DCD
attempted
DCD not
attempted
Mean SD Mean SD
P-
value
Total dose 17.2 29.2 11.6 18.2 0.13
WE GOT TO THINKING….
• Time has to play a role…
• Are there differences in patients who
die quickly than those that don’t
Morphine doses between DCD attempted vs. not attempted who died within 2 hrs of WLSM
Comparing total doses between DCD attempted vs. not attempted who died and did not died
within 2 hours of WLST
Did not die=
Died=
STUDY CONCLUSIONS
Results
• Small differences in in
extubation rates
• Potential Small
differences in opioid
dosing
Keep in mind…
• Small sample size-hypothesis
generating
• other sedatives!!
• protocolized WLSM
• Didn’t account for Drug kinetics
THE TAKE HOME SLIDE
• Time to death can be predicted using a combination of demographic,
physiological, physician prediction and WLSM approach predictors
• Approach to WLSM may be different between DCD and non DCD
patients- more research needed!
• This work further demonstrates the utmost importance of a
standardized and systematic approach to WLSM in all patients.
THANK YOU