Click Here To Download - Cancer Center Business Summit

voltaireblingΔιαχείριση Δεδομένων

20 Νοε 2013 (πριν από 3 χρόνια και 10 μήνες)

73 εμφανίσεις

Actionable Data
Analytics for
Cancer Care Re
-
Design


Tom Barr, MBA

Peter Kuhn, PhD

Debra
Patt
, MD, MPH

Disclosure
Information

Peter
Kuhn, PhD

Consultant
or Advisory Role:
Epic Sciences, Inc.

Stock Ownership
:
Epic Sciences, Inc.

Research
Funding:
Epic Sciences, Special Funded Project 1854,
“Characterization of Minority Cell Types in Peripheral Blood


Debra
Patt
, MD, MPH

Employment
or Leadership Position:
McKesson Specialty Health, Medical
Director, Healthcare Informatics





Please
note, all disclosures are reported as submitted to the Cancer Center Business Summit and are
available at cancerbusinesssummit.com.


Actionable Data
Analytics for
Cancer Care Re
-
Design


Tom Barr, MBA


The problem


The science


Examples of the science at work


Road Map

Clarion Call

Do the Right Thing

69.1% (1999)

1984

Cholesterol Screening

75.5% (2001)

1982

Mammography

48.1% (2000)

1981

Diabetic Eye Exam

53% (2000)

1977

Pneumococcal Vaccine

64% (2000)

1968

Flu Vaccine

Current rate of use

Landmark Trial

Clinical Procedure

Balas

EA, Boren SA., Managing Clinical Knowledge for Health Care Improvement. Yearbook of Medical Informatics 2000
.

Do
the

Right Thing

Quality of Pneumonia Treatment for Elderly, 2002
63.1
67.9
81
29.6
0
10
20
30
40
50
60
70
80
90
100
Received 1st
dose of
antibiotic within
4 hours of
hospital arrival
Received
recommended
antibiotics
consistent with
current
guidelines
Have blood
cultures
collected before
antibiotics are
administered
Received all
recommended
treatment
regimens
Measure
Percentage of patients
Source:
Centers for Medicare & Medicaid Services, Quality Improvement Organization Program,
2002.

81% of Medicare
pneumonia patients
get blood cultures
before antibiotics


68% get the right
antibiotics


63% get their first
antibiotic in a timely
manner


Yet, only 30% get all
of three
recommended
interventions

Healthcare & Biomedical IT is HARD


Individuals are highly variable biological systems.


Clinical measurements rarely have precise meaning.


Diagnoses lack clinical detail.


Clinical work is a chaotic, opaque ecosystem.


Perspectives vary by role.

Biomedical Informatics

Definition:

Science that deals with information, its structure, acquisition and use

Cornerstones:


Techniques to structure, discover, visualize & reason with information
content


Approaches to link people, process & technology together as a
system


Methods to evaluate systems and their technology components


Processes to facilitate change

Automation

Connectivity

Decision

Support

Data Mining

Mismatch between Computational Technique &
Scale of Problem

Data Mining

Automation

Connectivity

Decision
Support

Aggregate
EHR

Disease
management
dashboards

Work lists

Evidence
-
based
advisors


Match Computational Approach to Complexity
of Data

Stead WW. Electronic Health Records. In: Rouse WB,
Cortese

DA, eds.
Engineering the system of healthcare delivery
.
Tennenbaum

Institute
Series on Enterprise Systems, Vol. 3. Amsterdam: IOS Press; 2009.

Managing
Chaotic Events and Complexity


Work at multiple scales


Triangulate multiple signals for robustness

Satellite

Doppler Radar

Rain Gauge

R
E
P
O
S
I
T
O
R
I
E
S


transaction

processing


event life cycle

Holistic

Informatics


modeling &
simulation

Plan


optimization
algorithms


surveillance


visualization of
status

Activation

Task
Management

Informatics Foundation for Systems Approach to Care


Signal capture


edits


robotics

Task

Performance

Analysis

Granular

Automation

Evidence
-
based Medicine

Consistent Process

Visualization

of Results vs. Plan

Iterative

Improvement

Outcomes

Systems Approach in Practice

Congestive Heart Failure


Surveillance Algorithm


Integrated Decision Support

Confidential

System automatically clusters
diseases, drugs, and patients

More related

Less related

Complex
Care

Preventive
Care

Population Management Visualized


1000

Facts per Decision

10

100

1990

2000

2010

2020


Human
Cognitive Capacity

Need for Patient
-
Specific Decision Support Assistance

Structural Genetics:

e.g. SNPs, haplotypes

Functional Genetics:

Gene expression profiles

Proteomics and other

effector molecules

Decisions by clinical
phenotype

i.e., traditional health
care




Summary


Automation & connectivity get you only 60%


In addition to automation and connectivity need to view and leverage
clinical systems for Information Liquidity


Need for dashboards and knowledge engines that sit on top of the
information.


Need BI/Analytics Platform that leverage the information.


This architecture is scalable


Managing Risk:


What problems are worth solving (BI)


Key interventions to reliably intervene (dashboards + automation
of decision support)

Actionable Data
Analytics for
Cancer Care Re
-
Design


Debra
Patt
, MD, MPH

Who do actionable analytics serve?


Patients


Physicians


Systems of care delivery


Payers?

Actionable analytics can influence…


Quality


Errors


Compliance with EBM


CDST


Facilitate access to clinical trials


Identify novel treatment options


Support program compliance

How can analytics feedback to support
care re
-
design?


Feed back information to doctors


Variance


Regression to the mean


Feed back information to doctors


Expectations of care differ from reality


Identify opportunities to optimize treatment



…In this way they are all learning systems of
care

How can you facilitate compliance

with EBM?


Evidence based pathways


CDST at POS


Feedback reporting


Learning systems of care

Quality


Structure


Process


Outcome

Innovent Portal is a
web
-
based clinical
decision tool, which
recommends
suggested Pathway
treatment options
based on patient
-
specific criteria.

Pathways Decision Support Tools

Web Portal

Copyright © 2010 US Oncology, Inc. All rights reserved.

32

Reporting Tools



Tracks
monthly
performance




Progress
over time

Studies Conducted by USON Practices Show Value of
Pathways and Comprehensive Approach

| For internal use only/proprietary and confidential.


Quality

structure, process, outcome


Value
-

Outcome/Cost


Influencing compliance with programs


Expectations of care delivery


Identify target populations for
implementation


Feedback reporting to physicians at regular
intervals


Facilitate compliance with program
implementation

Universal Paradigm for

Quality Programs


Pathways


Advanced care planning


MU Certification


Clinical trial identification


Actionable Data
Analytics for Cancer
Care Re
-
Design


Peter Kuhn, PhD

Human Investigation, the science
of actionable analytics


In
-
patient, out
-
patient and
hom

care monitoring


Efficiency & Effectiveness


Personalized Medicine/Companion Diagnostics


High(
er
) predictive value of response


Predicting toxicity


Personalized Medicine: N of 1, who is N? How do we
understand N?

Translation between languages:

Cancer Center Business Language:

Accountable Care

Science Language:






Time as the Fourth Dimension

Personalized Medicine/Companion
Diagnostics


Past:


Biopsy from Melanoma: B
-
Raf wild type predicts
adverse effect, i.e. tumor progression


Biopsy from lung cancer: Eml4
-
ALK predicts 65%
response rate in 4% of patients


Now/Future


High
-
content liquid biopsy of the blood as
the

routine access to the disease


Quantitative protein and genomic marker
assessment

Personalized Medicine/Companion
Diagnostics


The GOOD


New, high precision analytical methods and instruments are
being developed for clinical utility (not just ‘translated’ from
research)


Higher data volumes allow for more accurate interpretation


The BAD


Cancer is a really hard disease that is constantly changing
especially under highly targeted therapy


The Ugly:


Only 62% overlap when the first 1000 genes of a RNA seq
experiment are compared by two different software packages


How does personalized cancer care become useful
in broad cancer center community reaching
greater than 80% of all patients?

Personalized
Medicine/Companion Diagnostics

PERSONALIZED MEDICINE: N OF 1, WHO IS N?
HOW DO WE UNDERSTAND N AND USE IT TO
PREDICT WHAT WILL HAPPEN TO 1?





Main features:



Nodes are web pages



Nodes are linked by directed edges



Knowledge on all edges



No knowledge on node %
(importance)







Google uses the internet model to:



Perform simulated internet searches:



Individual searches with random walks



`Ensemble’ searches with Monte


Carlo simulations


Run tests under different scenarios


Calculate `average’ number of steps from

node
i

to node j


Search: Lung Cancer

Google’s
PageRank

Algorithm:
A Markov Chain Model

43

1

1/2

1/2

3/4

1/2

1/2

1/2

1/2

3/8

5/8

1

1

1

1/3

1/3

1/3

Edge Weights:
`Transition’ probabilities

1/4

44

2
3
2
4
2
5
2
6
2
8
3
0
3
1
3
3
3
4
3
8
3
9
4
0
4
1
4
2
4
4
4
8
4
9
2
7
2
9
3
2
3
5
3
6
3
7
4
3
4
5
4
6
4
7
5
0
1
5
6
7
1
2
1
6
1
7
1
8
1
9
2
2
2
3
4
8
9
1
0
1
1
1
3
1
4
1
5
2
0
2
1
Lung

Prostate


27 sites


729 connections


Ave weight 0.037


Max weight 0.188


22 sites


462 connections


Ave weight 0.048


Max weight 0.275




Disease progression is

analogous to web
-
surfing

from site to site


Key parameters are the

transition probabilities
a
ij





45



First metastatic site as spreader or sponge

Spreader
:

Yahoo?





46

Lung self seeds, Prostate does not





Lung


Adrenal

Prostate


Bone

Spreader diagrams









48



Actionable Science Outcomes


It is the
combined
characteristics of the
primary and 1
st

metastatic site that largely
determines progression paths and timescales
of progression


Adrenal gland is a spreader in lung cancer


Bone is a spreader in prostate cancer


Lung cancer with Adrenal gland
metastasis could be a stage IIIb disease
resulting in adrenalectomy and curative
intent treatment

Prostate Cancer can change within 3 weeks


Take a blood sample every week (at
Walgreens?) and ship to central lab


Run Epic Sciences CTC test


Run Cansera Genomics test on identified cells


Shock and Awe at what the speed of response
of prostate cancer in the patient

1 Patient, changing profile under Rx pressure

3 Weeks of Targeted Therapy

confidential

51


Clone A

is
sensitive to
treatment


Clone C

is
resistant to
treatment


It took only 3 weeks
to select for
Clone
3

in the patient

Actionable Science


In CML, the ability to identify the first
emergence of the resistant clone was a game
changer


we are seeing this for the first time
in the carcinomas


Specific changes are evidence generation for
clinical trials to identify alternative therapies
and/or combination therapies


What’s Next



Regulatory and Reimbursement of the individual tests



Integrate multiple modalities such as PET/CT imaging
AND blood testing.



Enable science in the clinic with large numbers of
patients over time


This will benefit you because

Accountable Care = 4
th

Dimension