7/15/2015
TM
INCLUSION OF DATA-DRIVEN RISK
PREDICTIONS IN RADIATION TREATMENT
PLANNING IN THE CONTEXT OF A LOCAL LEVEL
LEARNING HEALTH SYSTEM
Todd McNutt PhD
Associate Professor
Radiation Oncology
Johns Hopkins University
Disclosures
This work has been partially funded with
collaborations from:
Philips Radiation Oncology Systems
Elekta Oncology Systems
Toshiba Medical Systems
as well as
Commonwealth Foundation
Maritz Foundation
2
Learning health system
Decision
Point
Controls
Outcomes
Facts
time
Decisions
Knowledge
Database
Facts
Presentation of
Predictions
Controls
Predictive
Modeling
Predicted
Outcomes
Data Feedback
(Facts, Outcomes)
3
1
7/15/2015
Mandible
vs
PTV_7000
pt: 300
7/15/2015
4
Mandible
vs
PTV_7000
pt: 822
7/15/2015
5
Mandible
vs
PTV_7000
pt: 295
7/15/2015
6
2
7/15/2015
Mandible
vs
PTV_7000
pt: 258
7/15/2015
7
Mandible
vs
PTV_7000
pt: 234
7/15/2015
8
Shape-dose relationship for
radiation plan quality
Shape relationship
DB of prior patients
Dose prediction
parotids
1
Right parotid
Left parotid
0.9
0.8
PTV
normalized volume
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
5 10 15 20 25 30 35 40 45 50 55 60 65 70
Dose (Gy)
For a selected Organ at Risk and %V, find the
lowest dose achieved from all patients whose
%V is closer to the selected target volume?
Decisions:
• Plan quality assessment
• Automated planning
•
IMRT objective selection
• Dosimetric trade-offs
3
7/15/2015
Interface
7/15/2015
10
Sample automated radiation
planning result
Original plan
Automated plan
1
30% reduction in dose to parotids
0.9
Auto plan
Original Plan
Dot: right
No-dot: left
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
10
20
30
40
Dose(Gy)
50
60
70
80
brain (Gy) (max)
Brainstem (Gy) (max)
Cord4mm (Gy) (max)
L inner ear (Gy)(mean)
original
61.25
54.58
41.75
57.18
re-plan
56.33
46.48
37.89
43.72
R inner ear (Gy) (mean)
mandible (Gy) (max)
larynx for edema (V50)
esophagus (Gy)(max)
original
40.57
66.58
61%
63.74
re-plan
38.38
63.78
59%
61
Current dose based auto-planning
• Has demonstrated improved quality
• Removed human variability for standard
cases
• Now advancing commercially
7/15/2015
12
4
7/15/2015
That was all DVH based
• Dose is not what matters to the patient
• Quantify the patient experience?
Should we just apply existing NTCP and
TCP models to dose predictions?
…or should we try to expand the knowledge
based approach using clinical data?
7/15/2015
13
# patients
Mucositis data collected at JHU
Toxicity grade
14
Promote Culture of Data Collection
Data collected over entire treatment
Consult
Weekly On
Treatment
Demographics
Diagnosis
Staging
Baseline Tox
Baseline QoL
History
Toxicity
QoL
Patient status
Symptom Mgmt
Simulation
Planning
Targets
OARs
OVH
Rx
Dose
DVH
Auto Risk
Plan Based
Image
Guidance
Motion
Disease
Response
Symptom Therapy
Mgmt
Mgmt
End of
Treatment
Acute toxicity
QoL
Patient status
Symptom mgmt
Disease response
Follow Up
Late toxicity
QoL
Patient status
Disease response
At what time point do we have
enough data to make decision
based on future prediction?
Input Variables => Prediction?
5
7/15/2015
Data Collection in Clinic
Quality of life
Clinical Assessment
Disease Status
FACT HN
SSQ
SHIM
IPSS
PAN26
7/15/2015
16
Head and Neck Inventory
~800 pts up to 6 yr follow up
7/15/2015
17
Head and Neck Inventory
500
Toxicities (24var/pt)
450
Quality of Life (64var/pt)
400
Measured data (6 var/pt)
350
Disease response(8var/pt)
300
250
200
150
100
50
0
0-2w
7/15/2015
2-4w
4-6w
6w-3m
3m-6m
6m-1y
1-2y
2-3y
3-4y
4-5y
5-6y
>6y
18
6
7/15/2015
Organs at risk with full 3D
dosimetry
600
500
400
300
200
100
0
7/15/2015
19
Prostate Inventory
~1700 pts - ~650 with dose
20
>6
7/15/2015
yrs
DVH, Toxicities and Grade
distributions
Dysphagia
Larynx_edema
30% Volume
Voice Change
Larynx
50% Volume
Toxicity Grade
0,1,2,3,4,5
Mean and stddev
of DX% at grade
Number of
patients by
grade at D50%
7
7/15/2015
Toxicity Prevalence
(P. Lakshminarayanan)
4 yrs
1 yr
1 yr
22
7/15/2015
Toxicity and Dose Volume Histogram
(Scott Robertson et al…)
23
Voice Change
8
7/15/2015
Bad DVH!
≠
• DVH assumes that every sub-region of an OAR has the
same radiosensitivity and functional importance to the
related toxicity
• DVH assumes that each OAR is uniquely responsible for
the overall human function related to the toxicity
Spatially dependent features of
dose in the structures (F. Marungo et al.)
Method
Voice dysfunction
Xerostomia
n=99, n+=8, n-=91
n=364, n+=275, n-=89
0.915
0.743
Bagged Linear Regression (1000
iterations)
0.905
0.737
Naïve Bayes
Linear Regression
0.900
0.896
0.734
0.731
Random Forest (1000 trees)
0.724
0.683
NTCPLKB
0.596
0.700
Bagged Naïve Bayes (1000 iterations)
Weight loss prediction
(N. Minoru, S. Cheng et al…)
Endpoint: > 5kg loss at 3 months post RT
At end of RT
At planning
no
yes
Able to eat foods I like >= 3
(FACT)
no
yes
oropharynx
tongue
nasopharynx
nasal cavities
Larynx D55 < 27.5Gy
Diagnostic ICD-9
Larynx
salivary glands
thyroid
hypopharynx
(CACAE)
Combo Masticatory
Muscle D41 < 38.4Gy
Larynx D80 < 23.0Gy
FALSE
FALSE
FALSE
Combo Parotid D61< 7.5Gy
Larynx D10 < 42.3Gy
(CACAE)
Nausea < 1
Distance: HD-PTV to Superior
Constrictor Muscle >= 1.1cm
FALSE
Diagnostic ICD-9
Skin Acute < 3
oropharynx
tongue
nasopharynx
nasal cavities
larynx D59 < 27.4Gy
Combo Parotid
D50< 13.5Gy
FALSE
Larynx
salivary glands
thyroid
hypopharynx
FALSE
TRUE
N stage < 2
Pain Intensity
- Current < 5
Diagnostic ICD-9
tongue
Distance: LD-PTV to
Larynx >= -1.3cm
Combo Parotid
D96< 6.5Gy
Cricopharyngeal
Muscle D100 < 38.1Gy
Esophagitis/
Pharyngitis < 3
(CACAE)
FALSE FALSE TRUE TRUE FALSE FALSE TRUE FALSE TRUE FALSE TRUE FALSE TRUE FALSE TRUE
7/15/2015
27
9
7/15/2015
Pancreas Resectability
Volume of Kidney (%)
(S. Cheng et al…)
100
0
-5-4-3-2-1 0 1 2 3 4 5 6 7 8 9101112131415
Distance from PTV (cm)
Variable, mean
LA (n=76)
BR (n=20)
P-value
Distantce_SMA_0%
-0.8302
-0.3216
0.0764
Distantce_SMA_25%
-0.3739
0.1231
0.0922
Distance_SMA_50%
-0.0362
0.4849
Distance_SMA_75%
0.4101
0.9975
0.0805
Distance_ClosestVessel_0%
-1.0421
-0.4121
0.0361*
Distance_ClosestVessel_25%
-0.6513
-0.0427
0.0454*
Distance_ClosestVessel_50%
-0.3894
0.2739
0.0373*
Distance_ClosestVessel_75%
-0.08
0.5603
0.0238*
PTV volume
89.2791
66.7585
0.0065*
0.0882
7/15/2015
28
Summary
• We can quantify the patient experience and are
improving our capabilities rapidly
• It is possible to collect and house RT data/knowledge
in a clinical setting
• Current dose based auto-planning utilizes a learning
health system
• Data science models are maturing that can convert
the knowledge to clinical predictions
• Incorporation of these predictions into the planning
process would make Leonard “Bones” McCoy proud
• The potential to have clinical impact is evident…
…we have work to do which requires real partnership
with our clinicians
Acknowledgments
•
–
–
–
JHU-RO
– Sierra Cheng MD
– Michael Bowers BS
– Joseph Moore PhD
– Scott Robertson PhD
– Pranav Lakshminarayanan
– John Wong PhD
– Theodore DeWeese MD
GI Team
– Joseph Herman MD
– Amy Hacker-Prietz PA
H&N Team
– Harry Quon MD
– Ana Keiss MD
– Mysha Allen RN
– Sara Afonso RN
Toronto-Sunnybrook
– William Song PhD
– Patrick
•
JHU - CS
–
–
–
Russ Taylor PhD
Misha Kazhdan PhD
Fumbeya Murango BS
•
Philips PROS
•
Toshiba
–
–
–
–
–
Karl Bzdusek BS
Minoru Nakatsugawa PhD
Bobby Davey PhD
Rachel-Louise Koktava
John Haller
•
Elekta
•
University of Washington
–
–
–
–
Bob Hubbell
Kim Evans MS
Mark Philips PhD
Kristi Hendrickson PhD
10
7/15/2015
Precision Radiotherapy
Treatment Planning
OVH: serial vs parallel
Target
OAR1
OAR2
r70,1
r70,2
For parallel organs, OAR2 is more easily spared.
For serial organs, OAR1 is more easily spared.
Problem
Ability to advance radiotherapy is limited by our
knowledge of which patients are at risk of high
grade toxicity or of limited ability to cure.
Knowledge from clinical trials is quite coarse and
fails to consider all of the aspects of the
individual patient.
‘Big Data’ offers an opportunity to better predict
treatment outcome and provide improved
clinical decisions for individual patients.
11
7/15/2015
Toxicity trends during and after
treatment – detect outliers
During Treatment
Follow up
Swallowing
Worsens after Tx
for many patients
then improves long
term
Mucositis
Heals after Tx for
most patients
Dysphagia
Inflammation
Xerostomia
Tends to be
permanent
Dry Mouth
#pts
Toxicity grade (0-5)
Oncospace Consortium Repository
(It’s all about the data)
Johns Hopkins
U. Washington
U. Toronto
Sunnybrook
U. Virginia
Institution X
Knowledge Base
Registry
Quality Reporting
$/pt
N
Decision Support
Research
What can we do with the data?
– Clinical (prostate, pancreas)
– Efficient high quality plan
parotids
PTV
DB of prior patients
Dose prediction
1
normalized volume
Shape relationship
• Shape based auto-planning
0.9
Right parotid
Left parotid
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70
Dose (Gy)
• Weight loss prediction
– Improved symptom management
• Toxicity Risk
– DVH based
– Spatial dose based
≠
• Disease response prediction
– Pancreas resectability
– Head and neck HPV dose de-escalation
36
12